Research On Target Tracking Based On Adaptive Occlusion Judgment And Multi-feature Fusion

Visual object tracking is an important research subject in the field of machine vision,and is widely used in many fields in industrial production,electrical engineering,and social life.At present,with the development of science and technology,although the performance and types of various tracking methods are increasing,there are still limitations in solving problems such as occlusion,illumination change,scale change and real-time tracking.Therefore,it has become a hot research topic to design a tracking algorithm that combines tracking effect and real-time performance in complex scenes.Based on this,the main research contents and achievements are as follows:(1)The basic principle of kernel correlation filtering tracking algorithm is analyzed.And according to the format of the existing public data set,a new data set is developed to increase the breadth of the test sample.The tracking error caused by camera distortion is eliminated by calibrating and correcting the monocular camera.(2)A multi-scale and multi-feature fusion target tracking method is designed.Firstly,the scale filtering mechanism is introduced to solve the scale limitation of KCF algorithm.Then,the multi-feature fusion and autonomous selection mechanism is designed to make full use of the feature information of the target and increase the recognition of the target.The mechanism extracts CN feature,HOG feature and LBP feature respectively,designs three feature fusion modes,and selects APCE as confidence score to evaluate the performance of the three fusion modes to determine the optimal fusion scheme.Experiments show that the designed target tracking method improves the precision and success rate obviously,and has good robustness in dealing with scale variation,deformation,blur and other aspects.(3)Aiming at the drift problem caused by fixed model update rate when the tracking target is occluded,the tracking method that can resist occlusion is studied,and the occlusion adaptive judgment mechanism and model updating strategy are designed.The response distribution of the target area during occlusion is analyzed.According to the analysis results,four evaluation functions are introduced to adaptively judge whether there is occlusion or not,and the degree of occlusion is defined using the double threshold mechanism.For the result of occlusion judgment,the weighted fusion method is used to apply the result to the update strategy of the model to realize the dynamic change of the update rate.The experimental results show that the proposed tracking method has outstanding performance in dealing with the occlusion problem and has obvious advantages in tracking speed.(4)In order to verify the rationality and practicability of the tracking method in this paper,an indoor mobile robot tracking system based on monocular vision is designed and developed.This system uses a radar car instead of a robot.Experiments are carried out in multiple scenes such as occlusion,illumination change and rotation,and good tracking effect and real-time performance are obtained,which proves that the tracking method in this paper has high practicability and rationality in practical application.