Research On Embedded Video Tracking Technology

The deep learning method is becoming more and more popular in the field of target tracking because of its powerful feature extraction ability.Among them,the target tracking algorithm based on Siamese network with high running frame rate and strong tracking fault tolerance rate has attracted the attention of many scholars.However,the Siamese network uses an offline method to achieve end-to-end model training,and selects the first frame as the tracking template.The trained model cannot change the network parameters and target template during the tracking process,which will cause the network to fail to detect changes in the appearance of the target in subsequent frames.Make real-time adjustments.During the online tracking process,when the target is deformed or interfered,because the algorithm cannot update the model,the target matching error or the target will be lost,and eventually the target tracking will fail.Based on the above reasons,this paper proposes the following improvements based on the Siam FC tracking algorithm:(1)A target-background comparison strategy is proposed,which is applied in the template branch to extract the effective information of the target in the first frame to the greatest extent.Thereby,the feature extraction ability of the network to the target is further improved;(2)After the search branch and the backbone network of the template branch,the channel attention and spatial attention modules are added at the same time,and the two attention modules learn which information to emphasize or suppress to improve the representation.(3)On the basis of the above,an improved Siamese network target tracking algorithm based on the hybrid attention mechanism is proposed.First,offline training is performed on the training set of the ILSVRC2015 dataset,and then it is trained on the ILSVRC2015 dataset.The algorithm was tested and evaluated with the test set of the aircraft category of the VOT2018 data set.From the experimental results,it is known that the success rate of the algorithm proposed in this paper is increased from 61.2%of the Siam FC basic algorithm to 68.4%,and the accuracy is increased from 77.1%.To 82.3%,the experimental results prove that the algorithm has good tracking performance.Aiming at the hardware implementation of target tracking algorithm,this paper proposes a video tracking scheme based on FPGA+HI3559 architecture,designs an embedded multi-core heterogeneous video tracking system with open space or sea and air as the background,and has a high level of video acquisition and display module,image storage The hardware circuit design and simulation of each functional module,MIPI communication module and HI3559 tracking module are carried out,and the experimental test is completed to verify the accuracy and real-time performance of the hardware architecture.