Research On Anti-Occlusion Pedestrian Tracking Methods Based On Correlation Filter

Pedestrian tracking is to estimate the corresponding state of the target in a subsequent video sequence based on the initial location and scale of a known pedestrian target.In recent years,pedestrian tracking technology is a research focus in the field of computer vision and artificial intelligence.It has achieved significant progress and has been widely applied in the intelligent transportation,intelligent monitoring,mobile robots,vision navigation and other fields.Since the proposed tracking method based on correlation filtering,it has been favored by a large number of target tracking researchers because of its excellent tracking accuracy and real-time performance.However,the occlusion problem often occurs in actual tracking scenes,and the correlation filter lacks corresponding countermeasures.It is easily interfered by the external environment,resulting in tracking failure.At the same time,scale change,background noise problem also increases the difficulty of tracking.Therefore,in this paper,the correlation filter tracking methods are improved in terms of anti-occlusion strategy,scale estimation,suppression of background noise,and corresponding solutions are proposed and experimental verification.The main contents of the paper are:First,aiming at the problem of pedestrian tracking with frequent or long-term occlusion in complex scenes,an anti-occlusion pedestrian tracking algorithm based on location prediction and deep feature rematch is proposed.Firstly,the occlusion judgment is realized by extracting and utilizing deep feature of pedestrian's appearance,and then the scale adaptive kernelized correlation filter is introduced to implement pedestrian tracking without occlusion.Secondly,Karman filter is introduced to predict the location of occluded pedestrian position,the deep feature and YOLOv3 is used to the rematch of pedestrian in the reappear process.Finally,simulation experiments show that the proposed algorithm can effectively detect and rematch pedestrian under the condition of frequent or long-term occlusion.Second,in view of the lack of the correlation filter's own mechanism to handle the scale change and occlusion of pedestrians,a scale adaptive transformation and anti-occlusion pedestrian tracking algorithm based on dual correlation filter is proposed.Pedestrian location prediction is first achieved by introducing a color probability model characterizing color histogram features in the kernelized correlation filtering framework.Secondly,the log-polar transformation strategy is used to convert the scale change to the translation change,and the translation estimation is combined with a weighted phase correlation method.Tracking confidence is also calculated by attaching a kernelized correlation filter,the model is updated normally at high confidence and stopped at low confidence to avoid introducing incorrect background information in the model.Simulation experiments show that the algorithm can effectively improve the effect of pedestrian scale transformation and occlusion on the tracking accuracy.Third,the correlation filter tracking method has excellent calculation speed due to the cyclic matrix and Fourier transform,but it also brings boundary effects.Intensive sampling of real background samples can reduce the boundary effect,but introduces more noise interference.Therefore,an objective function based on the filtering response and spatial constraints is proposed to optimize the filter,suppress the noise caused by aberrance response and excessive background information,and establish a more accurate tracker model.Periodic updating of the filter template is used to attenuate the interference caused by the occlusion of the target.Simulation experiment results show that the algorithm can effectively suppress the noise interference caused by dense sampling,has good anti-occlusion ability,and achieves better tracking accuracy and success rate in the test video.