Detection,tracking And Prediction Of Moving Targets Based On Deep Learning

In intelligent video analysis system,the detection and tracking of moving objects is a core link,and its performance directly affects the performance of action recognition,behavior analysis and other subsequent functions.Due to the complexity of the scene of video capture,the diversity of interference factors and the dynamic property of the moving object itself,the single frame image will be out of focus,blurred or lost and so on,which brings difficulty to the detection and tracking of moving objects.In this paper,we use computer vision technology and deep learning methods to carry out related research and exploration in the task of video moving target detection,tracking and trajectory prediction,and put forward an optimized dynamic neural network,which has achieved good results.The main work of this paper is as follows:Based on the analysis of the core components of the convolutional neural network,including convolution and pooling,the basic structure and main characteristics of SSD network in the target detection task are studied,and a new model two-tiny SSD is proposed to solve the problems existing in the moving target detection and tracking of SSD network.Based on SSD network,the two-tiny kernel phase processing mode is adopted to improve the detection and tracking of the targets effectively.By adjusting and setting the network structure parameters,the dynamic performance of the network is improved.Taking VOC datasets as the simulation test object,the results show that the accuracy,real-time performance,MAP and other aspects of target detection and tracking of two-tiny SSD have been improved.In terms of dynamic test,the two-tiny SSD network also achieved the desired effect.Experimental simulation on a subset of VOC data(human set)shows that the AP is significantly improved,and the detection and tracking of moving targets is more effective and reliable.By setting parameters such as confidence,the stability and accuracy of moving target detection and tracking are improved.By setting the distance between the center points between frames,the validity and persistence of detection and tracking are judged.The trajectory of the moving targets is predicted and judged dynamically,and the movement trend and dynamic range of the targets are estimated.The two-tiny SSD model proposed in this paper has achieved good results in both dynamic and static tests of VOC datasets,with high detection accuracy,fast detection speed,high MAP,strong tracking persistence,stable and reliable model performance,which has strong application value and engineering significance.