Over the past decade, continuous improvements and enhancements in CMOS image sensor technology have enabled it to move from primarily serving the low-end market to some of the most demanding high-performance

Moving target detection is a key technology in video image tracking and recognition systems. It has been widely used in video surveillance, traffic flow statistics, human-computer interaction, robotics and other fields. Currently, the commonly used methods include frame difference method, background subtraction method, optical flow method, etc. Among them, frame difference method and background subtraction method are suitable for detecting moving targets when the camera is stationary, and optical flow method can achieve better detection results when the camera is moving. In various application fields of moving target detection, there are high requirements for the real-time performance of the algorithm. Therefore, how to improve the operation speed to meet the real-time requirements is a problem that researchers need to solve.
Cellular neural network (CNN) is a mesh nonlinear circuit model with parallel processing capability [1]. Its basic unit is called a cell. The cell structure is simple and the cells are locally interconnected. Therefore, it is convenient to implement VLSI. The successfully developed CNN Universal Machine has been proven to have the computing power of a Turing machine [2]. As a neural network for integrated circuit implementation, CNN combines parallel computing and parallel structure well. It has a structure similar to the human retina. Therefore, it is of great significance to use CNN to explore visual computing and real-time image processing. At present, CNNs are mainly used in the field of real-time image processing, and have also been successfully applied in research fields such as pattern recognition, bionic eyes, autonomous robots, information security, and advanced brain functions [3-5]. There are many forms of CNNs, such as difference-controlled CNNs, fuzzy CNNs, multi-layer CNNs, and time-delay CNNs.
This paper first gives the basic concept of CNNs and analyzes its stability. Then, for the commonly used frame difference method and optical flow method for motion target detection, the implementation method based on CNNs is given. Finally, simulation verification is carried out using different video image sequences.

The CNN templates for operations such as thresholding, filtering, hole filling, edge estimation, and reverse selection can be found in references [7-8].
4 Optical flow motion target detection method based on CNNs
Motion generates optical flow, which is an approximate reflection of motion information. Motion detection based on the optical flow method uses the optical flow characteristics of the moving target that change over time, and detects the moving target by calculating the optical flow and segmenting the optical flow image. Since different objects in the optical flow field have different speeds, independent moving targets can be detected even when the camera is moving. The disadvantage of the optical flow method is that the calculation method is complex and the amount of calculation is large, so it is difficult to apply to occasions with high real-time requirements. This paper uses cellular neural networks with parallel computing capabilities to realize the estimation of optical flow fields.
4.1 Continuous time domain optical flow calculation description
Cellular neural networks perform information processing in the continuous time domain, so the continuous time domain description method of optical flow calculation is first considered. If the gray value of a pixel m in the image at time t is I(x, y, t), let the speed of point m be Vm=(u, v), then the Horn & Schunck optical flow calculation model, its optical flow vector is solved by the following set of equations:


4.2 Simulation test results
Take the highway image sequence to test the proposed optical flow motion detection method. The image sequence is taken under the condition of camera movement, and there are almost no static objects in the sequence. In order to obtain better detection results, after the optical flow calculation (calculation of motion vector amplitude), a series of operations such as filtering, thresholding, hole filling, edge detection, and noise extraction are used in sequence. The detection results of the CNNs optical flow method are shown in Figure 4. It can be seen from the simulation experiment that the proposed method can obtain correct detection results.

This paper explores the implementation method of cellular neural networks for commonly used moving target detection methods, and finally uses different video image sequences for simulation verification. The results prove the effectiveness of the proposed method.
References
[1] CHUA LO, YANG L. Cellular neural network: theory[J]. IEEE Transactions on Circuits and Systems, 1988, 35(10): 1257-1272.
[2] LINAN G, ESPEJO S, DOMINGUEZ C R. ACE4K: an analog I/O 64×64 visual microprocessor chip with 7-bit analog accuracy[J]. International Journal of Circuit Theory and Applications , 2002, 30(1): 89-116.
[3] BALYA D, ROSKA B, ROSKA T, et al. A CNN framework for modeling parallel processing in a mammalian retina[J]. International Journal of Circuit Theory and Applications, 2002, 30(2): 363-393.
[4] ARENA P, BASILE A, FORTUNA L. CNN wave based computation for robot navigation planning[M]. Proceedings of the 2004 International Symposium on Circuits and Systems, 2004: 500-503.
[5] PETRAS I, ROSKA T. Application of direction constrained and bipolar waves for pattern recognition[C]. Proceedings of the 6th IEEE International Workshop on Cellular Neural Networks and their Applications, Catania, Italy, 2000: 3-8.
[6] SLAVOVA A. Applications of some mathematical methods in the analysis of cellular neural networks[J]. Journal of Computational and Applied Mathematics, 2000, 114(6): 387~404.
[7] Ju Lei, Zheng Deling, Weng Yifang. Fast image segmentation method based on cellular neural network[J]. Journal of Beijing Technology and Business University (Natural Science Edition), 2005, 23(9): 32-34, 39.
[8] Ju Lei, Zheng Deling, Zhang Lei. Image filter based on difference control cellular neural network[J]. Journal of University of Science and Technology Beijing, 2005, 27(6): 375-379.