Research On Target Detection And Tracking Technology Based On Deep Learning

In recent years,object detection and tracking technology has become a new research hotspot and has received extensive attention in many fields.Target detection and tracking technology has challenges such as occlusion and mobile platform deployment in practical applications.This paper conducts research on these problems,mainly including the following work:Firstly,this paper proposes an improved kernel correlation filtering algorithm.The traditional kernel correlation filtering algorithm cannot effectively deal with the scenes such as the target being occluded and shaking,and is prone to tracking errors and missing targets.Tracking,based on the kernel correlation filtering algorithm,realizes the early warning of target occlusion through the analysis of the output map of the response value of the tracker,and predicts the target trajectory through the least square method,which makes up for the filter when the target is blurred or occluded.limitation.Using the OTB dataset for testing,the results show that the average tracking accuracy has increased from 58.6% to 86.6%.Since the least squares method is used instead of the filtering operation when the target is occluded,the tracking speed is also improved to a certain extent.Secondly,for the target detection task in common scenarios,this paper designs a target detection system based on an embedded platform.The YOLO V3 model is used as the main body of the target detection system.The data sets of common targets such as people and cars are used,the target detection model is trained and converted,and the target detection model is deployed by using the neural network inference check of the Hi Silicon processor to realize the detection function of common targets in the visible light scene.Finally,this paper uses FPGA KINTEX-7 and HI3559 A processor to develop an image processing board and supporting host computer software to realize the embedded platform deployment of target detection and tracking technology.A digital component serial interface is built through the FPGA platform,and the visible light camera is connected to realize the acquisition of video data.The algorithm of target detection and tracking is deployed on the Hi Silicon platform,and the communication with the host computer software is realized through the network port.The image data and the results of target detection and tracking are transmitted to the embedded platform at the same time as the switching commands of tracking and detection and the coordinates of the selected target.After the test of the embedded system,the hardware platform can realize the detection and tracking of infrared and visible light targets.Based on the embedded platform,it can realize the detection of common objects in visible light and infrared scenes,and can be selected by the host computer software to be tracked.object.In complex scenarios such as occlusion and shaking,the embedded platform can also effectively track the selected target.