Video Target Detection And Tracking Technology Based On Convolutional Neural Network

With the advent of national intelligence,video surveillance has become more and more important.Among them,target detection and tracking is the most challenging research direction in the field of computer vision.At present,many scholars have researched and proposed many excellent algorithms.However,in the actual application environment,target detection and tracking still have a series of difficulties and limitations.Due to the influence of some uncertainties during the tracking process,the tracking results will fail.Based on the theory of compressed sensing and the structure of twin networks,this paper studies the target detection and tracking of video images.The research contents mainly include the following three aspects.Firstly,this paper analyzes the application of compressed sensing theory and convolutional neural network in image processing,and also analyzes the advantages and disadvantages of object detection algorithms using general description operators.Secondly,this paper proposes an EKF?GOTURN target tracking algorithm based on CS theory and GOTURN algorithm.Firstly,the algorithm is used to compress the image by using Hadamard matrix as the measurement matrix.Then the compressed image is input into the neural network to train the network parameters.Finally,the extended Kalman filter is added to the twinning network structure to predict and locate the target search area.Through the benchmark dataset test,the results show that the algorithm has better speed and accuracy in the target tracking process.Finally,this paper proposes a tracking algorithm that uses ARPN target detection structure and self-update target of Siam-FC target tracking.The algorithm preprocesses the image first,then uses the ARPN structure to perform target detection,then inputs the recognition target to the Siam-FC production target template,and finally updates the target template every 15 frames during the tracking process.The experimental results show that the proposed algorithm has excellent performance in target tracking accuracy and can effectively resist the influence of interference factors.