Prevention and control tracking system based on 3D face recognition

Abstract

A prevention and control tracking system based on three-dimensional (3D) face recognition was designed to improve the target tracking accuracy of the prevention and control tracking system. The ARM control chip of TMS320DM6446 was selected as the control chip of the ARM control module. The CMOS image acquisition sensor of the image acquisition module collected face images. The collected images were transmitted to the 3D face recognition module. The 3D face recognition module used the Gabor wavelet algorithm to extract the 3D face contour features of the face image. Moreover, the LDA algorithm was used to recognize faces based on 3D face contour features. The 3D face recognition results were compared with the faces in the face library to determine whether prevention and control tracking were necessary. When prevention and control tracking was needed, the GPS tracking and positioning module embedded in the mobile device terminal of the target object was started. The GPS tracking and positioning module was used to prevent and control the tracking of the target. The results of prevention and control tracking were displayed to the system users using a VGA display. The experimental results indicated that the designed system could accurately recognize faces and achieve prevention and control tracking of the target based on the face recognition results.

Keywords

3D face recognition prevention and control tracking system control chip Gabor wavelet contour features LDA algorithm

1. Introduction

The rapid development of face recognition technology in artificial intelligence has increased the applications of face recognition technology in public areas. The prevention and control tracking system is currently widely used in public areas [1]. In these areas, this system uses image acquisition sensors to collect and record faces, and the obtained face information is compared with the face data in the system’s background [2]. The prevention and control tracking in public areas and quick recognition of faces can be realised through three-dimensional (3D) face recognition technology. The man-machine interface is critical in the prevention and control tracking system [3]. Given the efficient development of computer and information technology, man-machine interaction level has become an important development direction in this field. The efficient interaction between humans and computers can be achieved through man-machine interaction technology [4], making this interaction natural and friendly. In recent years, the acquisition performance of image acquisition sensors has also been continuously improved. The face information collected by image acquisition sensors [5] is analyzed and processed by information technology, which is crucial for improving the performance of the prevention and control tracking system. The efficient application of prevention and control tracking systems [6] greatly impacts social security and precision.

The 3D face recognition technology uses 3D data. This technology can avoid the influence of illumination, expression, and posture in face recognition [7]. It can also solve the low accuracy of two-dimensional (2D) face recognition. The image acquisition performance of a face image sensor is also crucial. An image acquisition sensor collects face images and improves face recognition accuracy [8]. The calibration of an image acquisition sensor can improve the lens distortion and the performance of 3D face recognition. This process enables the prevention and control tracking system to locate the face accurately and identify the target identity quickly, even in environmental noise. GPS technology exhibits extremely high positioning performance [9]. In 3D face recognition, GPS technology tracks targets [10] to realize the prevention and control of targets and track the required targets effectively.

Many researchers have studied the prevention and control tracking system. Ren et al. [11] designed a multifaceted tracking system. The system adopts clustering algorithms to cluster the collected face images. It also uses a double-tribrach twin network to achieve accurate face tracking for image clustering results. The system has high pedestrian tracking accuracy and a high face recognition rate. In the cross-camera environment, accurate face tracking can still be achieved. Although the system has high tracking performance, it has the defects of too complex computation and low tracking efficiency. Qi et al. [12] applied the MTCNN algorithm to face tracking. In their system, deep learning algorithms are combined with edge devices. Moreover, effective face tracking is achieved by using edge devices’ low power consumption and computing power. The system can improve the performance of face detection and tracking, bring high economic and social benefits, and solve resource shortages. Although the tracking system can realize face tracking and improve efficiency with edge devices, its tracking accuracy is low. Accurate face tracking cannot be achieved under fast target movement.

Given the problems in the practical application of the above tracking system, a prevention and control tracking system based on 3D face recognition was designed in the present study. It adopted a high-performance Complementary Metal-Oxide-Semiconductor(CMOS) image acquisition sensor to collect face images and extract 3D contour features contained in face images. According to the extracted 3D contour features, it realized the target prevention and control tracking. The system testing verified that the designed system exhibited high prevention and control tracking effectiveness for the efficient prevention and control tracking of targets. The designed system also possessed high applicability.

2. Prevention and control tracking

2.1 Overall structure of the system

The overall structure of the designed prevention and control tracking system based on 3D face recognition is shown in Fig. 1.

Figure 1.

Overall structure of the prevention and control tracking system.

In Fig. 1, the designed system mainly consists of a CMOS image acquisition sensor, an Advanced RISC Machines (ARM) main control module, an image acquisition module, a GPS tracking positioning module, a Synchronous Dynamic Random Access Memory(SDRAM) memory chip, and a Video Graphics Array(VGA) display. The CMOS image acquisition sensor was used to collect the face images to be recognized. The ARM control module adopted the image acquisition module to transmit the collected face images to the 3D face recognition module. After the 3D face recognition module completed the face recognition, the 3D face recognition results were compared with those in the face library to determine whether prevention and control tracking was needed. When the need for prevention and control tracking was determined from the results, the GPS tracking positioning module was started and used to track the target. The VGA display displayed the prevention and control tracking results to the system users. The system used an Inter-Integrated Circuit (I2C) bus configuration to configure the camera register of the CMOS image acquisition sensor. The format conversion module was also used to convert the SDRAM memory chip’s stored data and image data.

In Fig. 1, the designed prevention and control tracking system was based on face recognition technology to meet different applications’ target prevention and control tracking requirements. The 3D face recognition technology was slightly affected by environmental noise interference. It also exhibited high application performance. The collected face information could be used to recognize the target’s identity quickly, track the specific location of the target, and complete the prevention and control tracking of the target. Based on the prevention and control tracking system of 3D face recognition, the specific contents of prevention and control tracking of targets were as follows:

(1)

The image acquisition sensor of the image acquisition module was used to collect face images and extract the 3D face contour features of the collected face images. According to the 3D face contour features of the extracted face images, the Linear Discriminant Analysis (LDA) algorithm was used for 3D face recognition to recognize the face and obtain its identity information.

(2)

The GPS tracking and positioning module was activated to determine whether the target was a prevention and control tracking target and meet the demand for prevention and control tracking target by obtaining the specific location information of the person and realizing the prevention and control tracking of the target. The target’s position information was used to confirm the target’s historical trajectory, and the target’s position information was displayed at different times. The VGA display displayed the target’s position coordinates and basic information. The designed prevention and control tracking system displayed the target’s historical trajectory and position.

2.2 System hardware design

(1) ARM control module

The ARM control chip of TMS320DM6446 was selected as the control chip of the system control module. The chip had strong control and high digital signal processing ability [13], with the advantages of the Digital Signal Processor(DSP) control chip and ARM chip. The selected ARM control chip could achieve face recognition and efficient control of target prevention and control tracking. The chip could provide a codec framework, which could shorten the system development cycle [14], realize face recognition and target prevention, and control tracking. The RS232 serial port was selected as the ARM control module to connect the serial communication port between the upper and lower computers. This approach was convenient for viewing and controlling the Linux program developed by the system. The overall structure of the ARM control module is shown in Fig. 2.

Figure 2.

Structure diagram of the ARM control module.

Figure 2 shows that the ARM control module of the system selected the TMS320DM6446 chip as the control chip. The chip comprised a C64X $+$ core, ARM9 core, an on-and-off chip device, and an image processing unit. The chip adopted a dual-core architecture, effectively improving the system’s processing performance [15]. It supported the Reduced Instruction Set Computer (RISC) and DSP processor technology with rich user interfaces, maintained various operating systems, and exhibited high processing performance and low power consumption. The structure characteristics of the TMS320DM6446 chip used in the ARM control module are as follows:

1) ARM processor

The chip adopted the ARM9296EJ-S processor, which belongs to the ARM9 general-purpose microprocessor. It could provide a 16-bit Thumb instruction set and a 32-bit ARM instruction set for the system, with the advantages of low power consumption, small size, and high performance [16]. It could also meet the computation performance of the prevention and control tracking system. The ARM processor adopted a CP15 coprocessor to manage the instructions issued by the system and adjust the capacity of the system memory. The ARM processor contained many controllers, such as a power controller, a phase-locked loop controller, and an interrupt controller.

2) DSP processor

The DSP processor used the C64X $+$ DSP core to provide information processing functions. The operating frequency of the processor was 601 MHz. The processor used the pipeline operation [17] and an independent hardware multiplier to provide 4860 MIPS processing performance. The processor contained 128 universal registers with a size of 32 bits.

3) Image processing unit

The ARM controller set up an image processing unit. The image processing unit contained two configurable peripherals, the front and back end, used for the image input and output and for the display, respectively.

4) Memory organization and peripheral unit

The data and program memory management unit was used as the storage space of the ARM control module, including RAM, ROM, Double Data Rate 2 SDRAM (DDR2) memory, and corresponding external memory [18]. The external memory of the module was used to store data, codes, and other information about the prevention and control tracking system. The dynamic memory was used to save files, system cores, and application programs formed by the operation of the ARM chip. The secondary cache structure was selected to set the internal storage space of the ARM control module. The processor supported various serial interfaces, such as Universal Asynchronous Receiver/Transmitter (UART) interface, I2C interface, and audio serial interface, to achieve full-duplex communication.

5) Power unit

The power unit of the ARM control module comprised different types of power management modes to meet the low power consumption requirements of the prevention and control tracking system.

(2) Image collection module

The image acquisition module used a CMOS image sensor to collect face image information. The sensor integrates the peripheral circuit and the pixel array into the chip to form a complete image acquisition module. Compared with the face images collected by the Charge-coupled Device (CCD) cameras, the face images collected by the image sensor have price and power consumption advantages. The OV7725 CMOS sensor was selected as the face image acquisition sensor. It integrated a 1/4-inch single-chip VGA camera to meet the needs of face image acquisition. The sensor had low power consumption and low illumination. It also achieved high efficiency at 20–70^∘C. The OV7725 CMOS sensor had a photosensitive array size of 640 $\times$ 480, which could be configured with image characteristics, data format, and resolution to meet the needs of the prevention and control tracking system. In the prevention and control tracking system, the interface circuit structure of the CMOS image sensor and control chip is shown in Fig. 3.

Figure 3.

Interface circuit structure diagram of the image acquisition sensor.

The main features of the OV7725 CMOS sensor in the image acquisition module are as follows:

The sensor had good photosensitive characteristics. Moreover, it had good image acquisition performance even in low illumination and other environments.

The sensor used a standard Serial Camera Control Bus(SCCB) interface, which could configure and achieve good output for video streams in different formats, such as RGB and RAW Image Format (RAW).

The sensor supported transmitting different interfaces, such as VGA and Quarter VGA (QVGA). It could also collect face images with different resolutions.

The sensor had noise suppression and edge enhancement functions. It could also output high-definition and low-noise face images.

The sensor had frame synchronization mode and good automatic configuration performance.

The sensor had automatic white balance, automatic exposure, automatic band-pass filtering, and automatic black-and-white level calibration functions [19].

The sensor could automatically adjust the saturation, sharpness, and other parameters of images according to prevention and control tracking needs. It also had an automatic calibration function.

The I2C bus realized the efficient configuration of the image sensor camera by using the image acquisition module.

(3) GPS tracking positioning module

GPS tracking positioning module was an important part of the prevention and control tracking system. The GPS tracking positioning module selected the central control scheme to control the target’s prevention and control tracking positioning. The structure diagram of the GPS tracking positioning module is shown in Fig. 4.

Figure 4.

Structure diagram of the GPS tracking positioning module.

Figure 4 shows that the GPS tracking positioning module mainly included a Global System for Mobile Communications (GSM) unit, a GPS unit, and a power supply unit. The GPS9808 OEM chip was selected as the GPS chip of the tracking positioning module. The chip had excellent performance, compact structure, and high positioning accuracy. The GPS tracking positioning module used a GSM unit to realize the communication of tracking and positioning data. The GSM unit comprised a TC35i chip and a C8051F020-single chip microcomputer. The TC35i chip was a special chip that could realize GSM communication. It had a point-to-point communication function. It could also realize efficient communication in the GSM1800 frequency band. The GPS tracking positioning module used a power supply unit to provide a reliable power supply for the module. When the GPS tracking positioning module was running, the I/O serial port was used to output the positioning information of the target, such as longitude, latitude, time, and speed. The GPS tracking positioning module analyzed the positioning information of the prevention and control tracking targets from navigation messages [20]. It also displayed and stored the target positioning information in real-time through the display terminal.

2.3 Data extraction and management of face library

(1) Using a 3D face camera to extract face information of people in the park

The prerequisite for establishing a face library was to use a 3D face camera with specific parameters to collect the 3D face data of the target group. The depth accuracy of the 3D face camera was required to be $\leqslant$ 0.2 mm, the photographing time needed to be $\leqslant$ 0.1 s, and the best collection distance required to be within 75–85 cm. Moreover, it supported the manual entry of the ID card information and ID card intelligent card reader information to match the 3D face information with the personnel identity information. Finally, the shooting needed to be normal even when the ambient illumination was $\leqslant$ 220 LX.

Figure 5.

Collecting facial data using a 3D facial camera.

(2) 3D face data acquisition and management software

When the 3D face camera was used to collect and extract human face data, the corresponding 3D face data acquisition and management software should also be used to manage and edit face information and key information. Its main functions should include the following: face image viewing; support of the multiangle preview, zoom in, zoom out, and look back of the 3D face images after completion of face acquisition; real-time display of collected images; real-time display of the collected face point cloud and texture images.

Figure 6.

3D facial texture schematic diagram.

2.4 System software design

The system’s 3D face recognition module used the Gabor wavelet algorithm to extract the 3D face contour features in the face images. The LDA algorithm was used to realize 3D face recognition according to the extracted 3D face contour features. The 3D face recognition results were the basis for target prevention and control tracking.

(1) 3D face contour feature extraction based on Gabor wavelet

Feature extraction was a key part of 3D face recognition. Feature extraction was used to transform the 3D face data collected by the image acquisition sensor into feature data. Then, the separability of features was used to distinguish different faces. The contour line, a special curve in the 3D face model, could reflect the details of the face. Gabor wavelet was selected to extract 3D face features. According to the actual needs, the spatial and frequency sampling characteristics of the 3D face model could be adjusted to obtain the required 3D face contour feature data.

The one-dimensional Gabor function used in the Gabor wavelet algorithm is described as follows:

$\displaystyle g(x)=\frac{\|k\|^{2}}{\mu^{2}}\exp\left(\frac{\|k\|^{2}\|x\|^{2}% }{2\mu^{2}}\right)j\exp(\|k\|x)$ (1)

In Eq. (1), $k$ and $\mu$ represent the center frequency and bandwidth of the wavelet filter, respectively, and $j$ represents the complex operator.

$\frac{\|k\|^{2}}{\mu^{2}}$ and $\exp\left(\frac{\|k\|^{2}\|x\|^{2}}{2\mu^{2}}\right)$ are used to compensate for energy spectrum attenuation. They are also considered Gaussian functions.

The Gabor filter function for extracting 3D face contour features is complex. The real part of the complex function is described as follows:

$\displaystyle\text{Re }g(x)=\frac{\|k\|^{2}}{\mu^{2}}\exp\left(\frac{\|k\|^{2}% \|x\|^{2}}{2\mu^{2}}\right)\cos(\|k\|x)$ (2)

The imaginary part of the complex function is described as follows:

$\displaystyle\text{Im }g(x)=\frac{\|k\|^{2}}{\mu^{2}}\exp\left(\frac{\|k\|^{2}% \|x\|^{2}}{2\mu^{2}}\right)\sin(\|k\|x)$ (3)

The real and imaginary parts of the Gabor wavelet form oscillation near the edge of the 3D face image, including an unsmoothed peak response. At that moment, the extraction effect of the 3D face contour was poor. Gabor wavelet was used to retain only the corresponding amplitude. The retained amplitude reflected the energy spectrum of the local region in the 3D face image. The energy spectrum showed high smoothness around the real contour of the 3D face image.

In the 3D face image collected by the image acquisition sensor, the contour line in the middle of the face comprised $k$ points. The coordinate sequence set was represented by $\{(y_{0},z_{0}),(y_{1},z_{1}),\ldots(y_{k-1},z_{k-1})\}$ . Moreover, the contour points in the 3D face image were represented by the coordinate series $s(x)=[y(x),z(x)]$ . The contour points were all complex. The complex representation of the 3D face contour coordinate points is as follows:

$\displaystyle s(x)=y(x)+jz(x)$ (4)

The subcontour $s(x)$ in the face is regarded as a one-dimensional signal. The expression of the convolution processing for the contour lines by the Gabor filter is as follows:

$\displaystyle O_{\phi,v}(x)=s(x)\times g(x)$ (5)

In Eq. (5), $\phi$ and $v$ represent the filtering direction and filtering scale of the Gabor filter, respectively.

The set $O=\{O_{\mu,v}(x),x=0,1,\ldots,k-1\}$ obtained by the above process is the Gabor wavelet transform feature of the 3D face contour.

Gabor wavelet extracted the contour features of the 3D face images through the above process. The direction sensitivity of the Gabor wavelet improved the extraction of the required fixed direction change features. Gabor wavelet could obtain different types of local features of contour lines through multichannel filtering technology. It could also effectively extract facial contour features with only a small amount of data processing, meeting the real-time requirements of a prevention and control tracking system. Gabor wavelet had high extraction performance even in environments with illumination changes. Thus, it could improve the robustness of the prevention and control tracking system.

(2) 3D face recognition based on LDA

The LDA algorithm used the extracted 3D face contour features as the basis for face recognition. The sample set $z$ was constructed using the extracted 3D face contour features, and the LDA algorithm was used to recognize faces. The LDA algorithm was a supervised learning algorithm. When the LDA algorithm was used for classification, the class spacing should be maximized, and the intraclass distance of the classification should be minimized. For the problem of the 3D face recognition, $z$ was set on the 2D plane. Moreover, $z$ was a 2D vector determined by the horizontal coordinate $X$ and the vertical coordinate $Y$ . The obtained projection could effectively classify the original data set when $z$ was projected in the appropriate direction. The best projection direction $w$ obtained by the LDA algorithm was the 3D face recognition.

The LDA algorithm was used to determine the optimal projection direction $a$ . The process of 3D face recognition is discussed next in this paper.

During this process, the 3D face contour feature sample data were set as the M-dimensional vector. The number of positive and negative samples was $N_{1}$ and $N_{2}$ , and the number of positive and negative samples for training was $n_{1}$ and $n_{2}$ . The face recognition process is as follows:

(1)

The mean vector of the 3D face contour feature sample set is calculated as follows:

$\displaystyle M_{i}=\frac{1}{n_{i}}\sum{x_{k}}$ (6)

(2)

The interclass dispersion matrix of the samples in the 3D face contour feature sample set is represented as follows:

$\displaystyle\overrightarrow{Z_{i}}=\sum(-M_{i}x_{k}^{2}-Mi)x_{k}^{T}$ (7)

(3)

The total dispersion matrix in the sample classification within the sample set of the 3D face contour feature is represented as follows:

$\displaystyle\overrightarrow{Z_{w}}=\overrightarrow{Z_{1}}+\overrightarrow{Z_{% 2}}$ (8)

(4)

The interclass dispersion matrix of the samples in the 3D face contour feature sample set is represented as follows:

$\displaystyle\overrightarrow{Z_{B}}=(M_{1}-M_{2})\ast(M_{1}-M_{2})^{T}$ (9)

(5)

The projection matrix of the Fisher classifier is represented as follows:

$\displaystyle a^{\ast}=\overrightarrow{Z_{a}}^{-1}(M_{1}-M_{2}).$ (10)

(6)

The extracted 3D face contour feature data are projected into different categories in vector $a^{\ast T}X$ , with the largest discrimination.

Three-dimensional face recognition was a multi-classification problem. The number of categories in the face library corresponding to 3D face recognition was supposed to be $C$ . The 2D face could not search for an appropriate classification vector. The 3D face recognition was completed by increasing the number of projection vectors. A projection vector set was supposed to be $A=[a_{1}|{a_{2}}|\ldots|a_{k}]$ , and the element in the set was an n-dimensional column vector. The projection of the 3D face contour feature sample $x$ on the projection vector $a$ was represented by $y=a^{T}x$ . $u_{i}$ was used to represent the center of the face recognition category $i$ . The interclass hash formula of category $i$ is as follows:

$\displaystyle\overrightarrow{Z_{ai}}=\sum\limits_{x\in ai}{(x-u_{i})(x-u_{i})}% ^{T}$ (11)

The calculation formula for the center $u$ of all samples $x$ is as follows:

$\displaystyle u=\frac{1}{N}\sum{N_{i}u_{i}}$ (12)

In the 3D face contour feature sample set, the calculation formula of the interclass hash degree is as follows:

$\displaystyle\overrightarrow{Z_{b}}=\sum\limits_{i=1}^{C}N_{i}(u_{i}-u)(u_{i}-% u)^{T}$ (13)

The interclass distance $J(a)$ of multiclass targets was determined according to the data distribution obtained after projection. The interclass hash matrix obtained after the projection is as follows:

$\displaystyle\overrightarrow{Z_{a}}=\sum\limits_{i=1}^{C}\sum\limits_{y\in a_{% i}}\left(y-\frac{1}{N}\sum\limits_{\forall y}y\right)^{T}$ (14)

The formula of the interclass hash matrix obtained after projection is as follows:

$\displaystyle\overrightarrow{Z_{B}}=\sum\limits_{i=1}^{C}N_{i}\left(\frac{1}{N% }\sum y-\frac{1}{N}\sum\limits_{\forall y}y\right)\left(\frac{1}{N}\sum y-% \frac{1}{N}\sum\limits_{\forall y}y\right)^{T}$ (15)

The interclass distance of multiclass targets could be obtained by calculating the sum of the hash degrees of various centers relative to the full sample centers. The calculation formula is as follows:

$\displaystyle J(a)=\frac{\overrightarrow{Z_{B}}}{\overrightarrow{Z_{a}}}=\frac% {a|a^{T}\overrightarrow{Z_{B}}|}{|a^{T}\overrightarrow{Z_{a}a}|}$ (16)

According to the actual needs of 3D face recognition, the feature vector of the previous maximum eigenvalue was selected as the result of face recognition to obtain the feature vector in Eq. (16). Only one classifier was required to be trained when the LDA algorithm was used for 3D face recognition. This classifier could be used as the final classifier for 3D face recognition to reduce dimensionality. The algorithm had high classification performance even with a large amount of data.

(3) Process of prevention and control tracking based on 3D face recognition

In target prevention and control tracking using 3D face recognition results, the prevention and control tracking system should ensure accurate face recognition and keep tracking while following the movement of the target to meet the real-time requirements of the prevention and control tracking system. The flowchart of target prevention and control tracking through 3D face recognition is shown in Fig. 7.

Figure 7.

Flowchart of the prevention and control tracking system.

Figure 7 shows that after the initial operation of the system, the image acquisition sensors and other equipment were initialized. The trained LDA algorithm was used to perform 3D face recognition for the collected face images. If the recognition results needed prevention and control tracking, the 3D face recognition results were added to the prevention and control tracking queue to facilitate further prevention and control tracking of subsequent targets. Based on the extracted 3D face contour features, the LDA algorithm trained the samples in the training sample set, called the recognition function. This algorithm also confirmed the person’s information in the face library according to the corresponding labels of the recognition results. After the character was determined as the target to be tracked, the target in the video should be searched. The GPS technology was used to complete the target prevention and tracking.

3. Results

The designed system was applied to a university campus to verify the application performance of the designed prevention and control tracking system. Moreover, a high pixel-level COMS image sensor was selected to collect face images. The sensor could also clearly capture the information in the image when shooting a wide range of images. It had a high stability of image detection. The camera had a built-in angle sensor, which was convenient for efficiently calibrating the image acquisition sensor.

The system in this study calibrated the image acquisition sensor of face image acquisition. The calibration parameter settings of the image acquisition sensor are shown in Table 1.

Table 1
Calibration parameter settings of the image acquisition sensor

Index	Pre-calibration		After calibration
Focal length (px)	203.	52	202.	49
Focal length error (px)	1.	52	0.	49
Primary point (px)	125.	53	124.	58
Principal point error (px)	2.	53	1.	58
Distortion coefficient	0.	00564	0.	00154
Pixel error (mm)	0.	1584	0.	0856

Table 1 shows that after the parameter calibration of the image acquisition sensor for the face image acquisition, the acquisition error and distortion of the image acquisition sensor were significantly reduced. The experimental results in Table 1 proved that the system in this study could improve the accuracy of face image acquisition by calibrating the image acquisition sensor. Using the calibrated image acquisition sensor for face image acquisition helped improve the results of 3D face recognition and the system’s prevention and control tracking performance.

The face image samples collected by 1400 image acquisition sensors verified the acquisition performance of image acquisition sensors. The image acquisition error of the image acquisition sensor was counted. The statistical results are shown in Fig. 8.

Figure 8.

Statistical results of image acquisition errors.

Figure 8 shows that the image acquisition error was significantly reduced using the system in this study to calibrate the image acquisition sensor. This system used an image acquisition sensor in collecting face images to meet the needs of 3D face recognition and target prevention and control tracking. It also calibrated the projection matrix of the image acquisition sensor to correct the calibration results of the image acquisition sensor and improve the acquisition performance of face images.

Moreover, the system in this study used the calibrated image acquisition sensor to collect face images, as shown in Fig. 9.

Figure 9.

Image acquisition results.

Figure 9 shows that the system in this study could effectively use the image acquisition sensor to collect face images. These collected face images had high resolution. The face images collected by the high-resolution image acquisition sensors could effectively improve the performance of 3D face recognition and target prevention and control tracking. They could also provide an effective image basis for the prevention and control tracking system.

Aiming at the face image results collected by the image acquisition sensor, it extracted the 3D face contour features. The 3D face contour features extracted by the system in this study are shown in Fig. 10.

Figure 10.

Results of the 3D face contour feature extraction.

Figure 11.

3D face recognition results.

Figure 12.

Facial recognition management system.

Figure 10 indicates that the 3D face recognition module of the system in this study used the Gabor wavelet algorithm to extract the 3D face contour features effectively in the collected face image. The 3D face contour features extracted by this system contained rich information that could recognize face identity. The 3D face contour feature extracted by our system was an important basis for 3D face recognition. The larger the useful information contained in the 3D face contour feature was, the better the 3D face recognition performance was.

The system in this study was used to perform 3D face recognition on the collected face images. The face recognition results and the facial management system are shown in Figs 11 and 12, respectively.

Figure 13.

Tracking results of target prevention and control.

Figure 14.

Tracking and recognition accuracy of the system.

In Figs 11 and 12, the tracking results indicated that the system in this study could accurately recognize faces in images. This system could effectively extract the 3D face contour features from the face image. The extracted 3D face contour features were used as the basis of 3D face recognition. They accurately recognized the faces according to the extracted 3D face contour features. The ambient light of the face image in Fig. 11 was relatively dark. However, our system could achieve accurate face recognition with high applicability even in a dark environment.

The 3D face recognition results in Figs 11 and 12 were set as the basis of target prevention and control tracking of the prevention and control tracking system. The system was used to track the target. The moving trajectory of the target is shown in Fig. 13.

In Fig. 13, the tracking results indicated that the system in this study could determine the identity information and location information of the target based on the 3D face recognition results. This system used the GPS tracking positioning module to track and locate the target person and display the moving track of the target person. Figure 13 verifies that our system could effectively locate targets and achieve target prevention and control tracking. These features were suitable for target prevention and control tracking in subways and other public areas.

Our system was adopted to track the target for prevention and control. The tracking and recognition accuracy of the system when the target was moving at different speeds is shown in Fig. 14.

Figure 14 shows that the system could accurately track and identify the target when the target was moving at slow, medium, and fast speeds. Under different circumstances, the tracking and recognition accuracy of the system in this study was higher than 99%. The experimental results proved that this system could effectively track and identify the target accurately based on the 3D face recognition results. Our system could adapt to the target prevention and control tracking under different environmental changes and moving speeds. This system could also track and identify efficiently even when the target was moving rapidly, thereby meeting the prevention and control needs of different applications in parks and public areas.

4. Conclusion

The 3D face recognition technology had the advantage of high recognition accuracy. It improved the defects of the 2D face recognition technology, which was easily affected by noise and environmental changes. The 3D face recognition technology used image acquisition sensors to collect target face images and improve the accuracy and rapidity of the 3D face recognition, which had high promotion value. The prevention and control tracking system based on 3D face recognition was designed to realize the prevention and control tracking of the target. The designed prevention and control tracking system could display the target location information and historical trajectory. The experiment verified that the designed system could use 3D face recognition technology to achieve target positioning and tracking. The system had high stability and good target-tracking performance. The face image was collected using a calibrated image acquisition sensor to compensate effectively for the image acquisition error caused by camera lens distortion. Our system could improve the prevention and control tracking performance by taking advantage of the solid antinoise ability, high robustness, and wide application range of the 3D face recognition technology.

Footnotes

Acknowledgments

This study was supported by the Science and Technology Research Program of the Chongqing Municipal Education Commission (Grant No.KJQN202204014).

References

Gao

Zhao

Chen

. Multi-dimensional data modelling of video image action recognition and motion capture in deep learning framework. IET Image Processing.2020; 14(7): 1257-1264.

Wan

Chen

Bai

. Joint feature extraction and classification in a unified framework for cost-sensitive face recognition. Pattern Recognition.2021; 115(10): 107927-107933.

Trigueros

Hartnett

. Generating photo-realistic training data to improve face recognition accuracy. Neural Networks.2021; 134(9): 86-94.

Xiao

Xie

. Application of novel gabor-dcnn into rgb-d face recognition. International Journal of Network Security.2020; 22(3): 534-541.

Baltanas

Ruizsarmiento

Gonzalezjimenez

. Improving the head pose variation problem in face recognition for mobile robots. Sensors.2021; 21(2): 659-667.

. Research on occluded face recognition based on low-rank sparse and network learning. Computer Simulation.2020; 37(10): 14-17+96.

Kadum

. Solving optimization problem in face recognition based on filtering and face position. Solid State Technology.2021; 63(2): 513-522.

Arnia

Oktiana

Saddami

Munadi

Pradhan

. Homomorphic filtering and phase-based matching for cross-spectral cross-distance face recognition. Sensors.2021; 21(13): 4575.

Curtidor

Baydyk

Kussul

. Analysis of random local descriptors in face recognition. Electronics.2021; 10(11): 1358-1364.

10.

Gunawan

Halimawan

Suharjito

. Lightweight end-to-end pose-robust face recognition system with deep residual equivariant mapping. Procedia Computer Science.2021; 179(2): 648-655.

11.

Ren

. Multi face real-time tracking system based on DTN in multi camera field of view. Laser & Optoelectronics Progress.2022; 59(02): 19-129.

12.

Zhuo

. Low-power edge AI face detection and tracking system based on improved MTCNN algorithm. Application of Electronic Technique.2021; 47(05): 40-44.

13.

Liu

Yang

. Tracking control of modular multilevel converter based on linear matrix inequality without coordinate transformation. Energies.2020; 13(8): 1978-1981.

14.

Zhang

Jiang

Chen

Xiao

. Numa-aware dgemm based on 64-bit armv8 multicore processors architecture.Electronics.2021; 10(16): 1984-1988.

15.

Mazumdar

Basu

. Hardware emulation of energization of a long transmission line by high-frequency power electronic converter. IEEE Transactions on Power Electronics.2020; 35(9): 9267-9280.

16.

Esfahan

Soleimany

Aref

. Enhanced cache attack on aes applicable on arm-based devices with new operating systems. Computer Networks.2021; 98(8): 108407-108411.

17.

Rahman

Paul

Rana

Rahman

. Performance evaluation & operation of dsp control ups system. Engineering.2020; 12(9): 666-681.

18.

Zhu

. Design of control system for adaptive front lighting system. Electronic Design Engineering.2020; (21): 125-129, 135.

19.

Zhang

. Intelligent vision measurement of small defects in high precision gear. Ordnance Material Science and Engineering.2021; 44(01): 83-87.

20.

Chen

Tan

Cao

Fan

. Modeling and improving the energy performance of gps receivers for location services. IEEE Sensors Journal.2020; 20(8): 4512-4523.

Prevention and control tracking system based on 3D face recognition

Abstract

Keywords

1. Introduction

2. Prevention and control tracking

2.1 Overall structure of the system

(1) ARM control module

1) ARM processor

2) DSP processor

3) Image processing unit

4) Memory organization and peripheral unit

5) Power unit

(2) Image collection module

(3) GPS tracking positioning module

(1) Using a 3D face camera to extract face information of people in the park

(2) 3D face data acquisition and management software

(1) 3D face contour feature extraction based on Gabor wavelet

(2) 3D face recognition based on LDA

(3) Process of prevention and control tracking based on 3D face recognition

Table 1 Calibration parameter settings of the image acquisition sensor

Footnotes

Acknowledgments

References

Table 1
Calibration parameter settings of the image acquisition sensor