Research On Target Tracking Algorithm In Intelligence Video Monitoring

Intelligent video surveillance technology has made people's life more convenient especially in improving the quality of service industry,helping civil servants retrieve past information,guaranteeing people's life safety and other fields in current era of pursuing efficiency.Object trackers have come a long way in intelligent video surveillance for the past few years.But targets are subject to interferences such as light intensity,occlusion,background diversity,and pose changes,which make the performance unstable.Human's ability to process information can quickly identify and track the objects with changes of scale,rotation and occlusion in complex scenarios.Therefore,tracking methods based on human vision have recently emerged.Correlation filter-based trackers have achieved superior performance for tracking,showing high precision and frame rates.There are many issues limiting the performance,such as the single method of image representation,repeated calculation of scale estimation and less robustness to complex scenes.In order to improve the accuracy of trackers and the ability to cope with challenging scenes,the research subject is about the research of target tracking methods in intelligent video surveillance.The work is mainly carried out from the following aspects:1)We presented a framework to train tracker with orientation gradient and color histogram features in translation and predicted the target size utilizing the kernel scale correlator.The model was updated online sparingly,alleviating the accumulation of small errors.2)By combining with the memory information processing mechanism,we proposed a tracking model and a correlation filter-based calculation model.The tracker model parameters and image features in the long-term memory space were replaced by the tracker model parameters and image features with high confidence,so the above framework was optimized.3)By comparing the methods that use different thresholds to judge the results' reliability,we can find out the best way to update the model with high confidence.So we can improve the robustness of the algorithm against challenges such as shape changes,scale transformations,motion blur and background clutter.Figure 28;Table 13;Reference 56...