Privacy-Oriented Successive Approximation Image Position Follower Processing

In this paper, we analyze the location-following processing of image by successive approximation with need for directed privacy. To solve detection problem moving human body in dynamic background, motion target module integrates two ideas feature information and model segmentation combines deep learning framework to complete detecting points key parts body. The depends on pose estimation algorithm, so research paper is based bottom-up multiperson method; firstly, all are detected extraction through convolutional neural network, then accurate labelling achieved using heat map offset fusion optimization method point confidence prediction, finally, results obtained. study correlation HOG KCF algorithm scale filter DSST form a principle MOSSE filter. solves problems lack low real-time rate improves tracking accuracy while ensuring performance algorithm.