Adaptive Background Aware Correlation Filter Tracking Method Research

Target tracking technology is an important research content in the field of computer vision.It has been widely used in video surveillance,human-computer interaction,military reconnaissance,precision strike,intelligent transportation,augmented reality,and medical diagnosis.In recent years,due to the strong performance of the related filter tracking method,it has become one of the research hotspots in today's tracking methods.In this paper,we study the adaptive contextual background-aware correlation filter tracking method.Proposed the background-aware correlation filter based on Kalman filter(KDCF),adaptive Background-Aware Correlation Filtering Tracking Based on Kalman Filtering(ADCF)and adaptive Context-Aware Correlation Filtering Tracking Based on Kalman Filtering(KCAF).The main research contents of the thesis and the new progress made are as follows:1.The paper proposes a background-aware correlation filter tracking method based on kalman filter focuses on the location selection of the background region in the context-aware correlation filter tracking method.The target motion state is estimated by kalman filter and the context of background information is acquired near the estimated target location to learn by the filter.The target location is still the image location corresponding to the peak value in the response map.The filter is more discriminative than others.The simulation results show that the proposed new method has strong robustness.2.The paper proposes an adaptive control parameter fine-tuning method focuses on context-aware correlation filter tracking method and the weight allocation problem of background regional training.To forecast the direction of target motion through linear estimation theory.It also gives more weight to the context of the target movement direction,and then to the deformation of the target and the internal and external rotation of the plane.The background region in the direction of the non-target movement still gives the equivalent weight.The difference between the target and the background area is improved effectively.3.This paper proposes a new blend of the location update strategy of adaptive scale is estimated and the scale is estimated focuses on the correlation filtering tracking method,which cannot adapt to the problem of target scale change.Through a new cropping mechanism,the weights of different sizes of scale samples are cut from the basic samples to the position filter learning.And the label matrix of the scale sample is given a new definition.The problem of target scale change is solved effectively.4.In the target tracking process,partial occlusion of the target or full coverage leads to the tracking drift and even the missing problem,combining with a new occlusion indicator APCE.Effectively address the problem of occlusion of the target.This paper conducts experimental verification and performance evaluation of the thesis algorithm and compares it with some current mainstream algorithms through the common database CVPR2013 Benchmark.The simulation results show that,the proposed algorithm is superior to other tracking algorithms on both precision and success rate.