Multi-target tracking based on target detection and mutual information

The key to achieving target tracking is to completely segment the target, reasonably extract features and accurately identify the target. At the same time, it is necessary to consider the time of algorithm implementation to ensure real-time and accuracy. This paper uses the Faster RCNN network structure to unify the extraction and classification of candidate detection areas into a single convolutional network architecture, and uses multi-person learning mechanisms to complete the learning of network parameters. while improving the efficiency and accuracy of physical detection. By using mutual information matching, in image registration, when two images are registered in a spatial position, the mutual information of the gray levels of the pixel pairs corresponding to the overlapped parts reaches the maximum value, and the best matching point of the image can be determined accordingly To achieve the effect of accurately tracking multiple targets. Combining three indicators of target tracking, average overlap expectation (EAO), tracking accuracy and robustness, the tracking algorithm proposed in this paper has the best effect on target tracking and has broad application prospects.