Research On Improved Multi Filter Long-term Target Tracking Algorithm

Moving target tracking has a very broad application prospect in military,intelligent transportation,intelligent video monitoring,human-computer interaction and other fields.In the practical application environment,the problems such as the change of target appearance,background interference,target occlusion and target out of view affect the tracking accuracy.The efficient-convolution-operator tracking algorithm simplifies the target model and greatly improves the tracking efficiency,but the single filter of the algorithm is difficult to adapt to the background interference,target occlusion and other complex environment.Besides the target can not be found after losing.To solve these problems,an improved multi filter long-term target tracking algorithm is proposed.The specific work is as follows:To solve the problem that a single filter is difficult to adapt to the complex changing environment,an improved efficient-convolution-operator tracking algorithm based on spatial-temporal regularization and consistency check analysis is proposed.Among them,the spatial-temporal regularization filter introduces temporal regularization into the loss function of correlation filter to solve the problem of target appearance change,and the consistency check filter reduces noise interference by reverse positioning method.The target features convolute with the spatial-temporal regularization filter,the consistency check filter and the efficient convolution operator filter respectively.The filter detection score with the largest peak-to-side ratio is used to determine the target location.Experimental results show that the improved algorithm can greatly adapt to the complex environment in the tracking process.In addition,an improved long-term target tracking based on efficient-convolutionoperator is proposed.That solve the problem of occlusion and out of view target tracking failure.Firstly,the occlusion detection algorithm is proposed to judge whether the current target is occluded or not,and stop the scale detection of occluded target to ensure the accuracy of scale model.Secondly,the judgment function of the target model is introduced,and the model of the unreliable target is not included in the update sequence of the appearance model,which greatly guarantees the accuracy of the model.Finally,the module of target redetection,which is composed of space distance weight,target detection and best partner similarity score,is used to relocate the out of view and lost targets.The experimental results show that the algorithm has the ability of anti-occlusion,can relocate the out of view and the lost target,and has high practical application value.