Real-Time Tracking of Dynamic Targets in Smart Cities with the Application of Artificial Neural Networks

In the public environment image of smart city, there are many “pain points” in the initial frame, and the dynamics of variable targets expand the degree of target tracking drift and limit the real-time tracking effect of dynamic targets. Based on the continuous frame dynamic target video image, the grayscale processing of the smart city public environment image is carried out, and a series of frame image preprocessing operations such as equalization, filtering and sharpening are used to strengthen the details of the dynamic target tracking domain of the smart city. Based on the artificial neural network framework, a twin network structure fused with convolutional neural network is constructed, combined with the dynamic object tracking rectangular box, the average offset of the target is controlled, and the continuous frame mask in the target tracking domain is output to generate a rectangular box tracking map, so as to realize the real-time tracking of dynamic targets in smart cities. The experimental results show that in the test, the minimum value of the average offset from the center position of the target obtained by the design method is 15 pixels, and the maximum value of the average processing delay of multiple frames is 0.15 ms, which indicates that the center position from the dynamic tracking target can be accurately calculated by using the design method, and the target tracking drift degree is light, which has a good real-time tracking effect.