Research On Pedestrian Tracking Technology Based On Siamese Network

With the rapid development of artificial intelligence,object tracking technology in the field of computer vision has become an important research topic.Although the research of target tracking has the input of many scholars and the output of research results,it is still a challenging task in the field of vision because of the complexity of the realistic scenes and application requirements.In many real scenes,the target to be tracked usually has the characteristics of deformation,occlusion or interference of similar objects,which has a great impact on the performance of the target tracking algorithm.For this problem,the following studies are carried out:(1)The improved lightweight network is used as the backbone network of feature extraction,and the deeper and lightweight network is used to extract features with more expressive ability,which is also suitable for the practical application of deep learning in the real environment.The experimental results show that,compared with the siamfc algorithm,the accuracy and success rate of OTB50 and OTB100 data sets are improved by using the lightweight feature extraction backbone network,the average accuracy and success rate of the two data sets are 75.1% and 55.8%;(2)On this basis,in order to obtain more favorable features of the model,a global channel joint spatial attention module is designed to further operate the extracted features to enhance the discrimination ability of the network model and emphasize the favorable information while restraining the redundant information,so that the algorithm can be better applied to the actual scene.At the same time,in order to obtain more accurate tracking results,in the generation of similarity score response graph,the cross-correlation results of the feature vectors of different parts are used as weighted average to improve the performance of the model.The experimental results show that the proposed algorithm performs well on the OTB series data sets,and the average accuracy and success rate of the two data sets,OTB50 and OTB100,reach 78.5% and 58.3%.Target tracking technology has more and more applications in the field of road traffic safety monitoring,robot and human-computer interaction,which has important research significance for the advancement of science and technology development and the expansion of civil safety.Target tracking technology also provides a research basis for higher level applications such as behavior identification and behavior analysis.