Research And Implement Of Trace,Detection,Recognition Algorithm On Moving Target

The computer vision is a kind of artificial intelligence technology which uses camera,computer,collector and so on to measure,track and recognize the object instead of the human eye,and does processing and analysis for the object.With the development of society and science,the research for video object tracking,detection and recognition has become a hot point in artificial intelligence field.However,common object recognition and detection algorithm based on deep learning is for single object label,it is opposed to the reality that most objects must have many labels.Therefore,this thesis analyses and studies major achievements of object tracking,detection and recognition in recent years in the field of deep learning,and proposed a comprehensive recognition algorithm of multi-label based on moving object.Firstly,this thesis analyses and studies the basic principle of convolutional neural network and the effectiveness and superiority of CNN in image feature extraction.And then implement an object appearance recognition network based on CNN.The principle of recurrent neural network and long short-term memory networks,and the advantages of LSTM to RNN are studied at the same time.This thesis implements an object action recognition algorithm with LRCN construction which is built with LSTM and CNN.And then,this thesis studies the related content of TLD target tracking algorithm,and improves the TLD tracking algorithm by use CNN as the detection and do experiments to verifies the effectiveness and robustness of the algorithm.At last,this thesis summarizes the research results of CNN,LSTM,TLD tracking algorithm and Softmax classifier,proposes a comprehensive recognition algorithm of multi-label based on moving objects combined with the multi-label learning.The algorithm can recognize objects with two aspects:object appearance and behavior.In this thesis,the function and design of each module are described in detail,and the basic structure and optimization operation of the algorithm model are explained.Also,the algorithm is tested and the results show the superiority of our algorithm compared to single-label recognition algorithm from function and performance.