Research On Target Tracking Algorithm Based On Deep Learning In Airport Video Surveillance

Video target tracking has been attracting attention in the field of computer vision,and it has been widely used in the fields of precision guidance and driverless driving.With the rapid development of the civil aviation industry,intelligent video surveillance has increasingly become an important auxiliary means of airport security.Targets in surveillance video often encounter disturbances such as occlusion,scale changes,and illumination changes,and the posture and speed of the target motion often change.The traditional tracking algorithm only pays attention to the apparent features of the target without considering the deep structure and geometric layout of the target.It has insufficient description of the target,and often leads to tracking failure in complex tracking situations.With the rapid development of deep learning in recent years,it has also been successfully applied in the field of target tracking.In this paper,the deep network hierarchy is complex and requires a lot of offline training,which can not realize the shortcomings of online update.The structure of the deep network is studied.The correlation between video frames is used to simplify the deep network structure and realize online feature extraction.Online tracking strategy to achieve adaptive robust tracking.The main research contents of this paper are as follows:By studying the tracking based on supervised convolutional neural networks,the supervised convolutional neural network(CNN)structure is more advantageous in processing images,depending on its unique network structure: convolutional layer,pooling layer Fully connected layer.The convolutional layer and the pooled layer are alternately connected to make the extracted features extremely robust.This is also the prospect of convolutional neural networks in the field of target tracking.Therefore,this paper is based on the CNN network structure and is based on depth.The learning target tracking algorithm is studied.Aiming at the complexity of existing CNN network hierarchy and the need of complex time-consuming offline training,a simplified shallow convolution feature learning network is studied by using video sequence inter-frame correlation,so as to achieve online tracking.The network generates a convolution filter suitable for the current tracking video frame by using the target information in the first frame and the background information of the adjacent frame,and simplifies the feature expression by using the sparse representation strategy.Through a simple two-layer convolution network,learn the characteristics suitable for tracking video and achieve online tracking.Experiments show that the features extracted by the network structure have better performance fortracking problems such as scale variation,background clutter,and mild occlusion;however,for fast motion,illumination changes,and severe occlusion,tracking failure may result.For the problem that the untrained convolution network can not achieve robust tracking under severe occlusion,illumination changes,and fast motion,the color feature is added to the convolution feature to enhance the apparent expression ability of the target,and in the tracking strategy.The motion information is introduced to realize that the candidate samples can achieve adaptive distribution according to the state trend of the target,thereby improving the robustness of the tracking.Experiments show that the optimized algorithm has been improved in terms of tracking effectiveness and success rate.Finally,the algorithm of this paper is simulated and implemented on MATLAB platform,and the effectiveness and practicability of the target tracking algorithm studied in this thesis are verified.