Tracking Algorithm Based On Improved Spatio-temporal Context Learning Algorithm

The rapid development of science and technology,people are becoming more aware of security.,intelligent tracking and monitoring systems are also more widely used in our lives.Target tracking technology has gradually become a hot spot for scholars to study,The Spatio-Temporal context algorithm has been widely concerned in recent years because it has fast tracking speed,strong robustness,and a certain degree of anti-occlusion ability.However,the algorithm will have problems such as tracking failure when the camera shakes or has severe occlusion.Based on previous researches,this paper proposes two improved algorithms based on space-time context algorithm:In order to eliminate the tracking failure caused by camera jitter,an improved space-time context algorithm based on composite feature extraction is proposed.This method is aimed at the characteristic that the algorithm can not automatically detect the first frame,Automatic detection of first frame with Adabost algorithm,Introduction of two more targeted Haar-like features,the improved Adabost algorithm can greatly enhance the detection rate,using the Harris-Surf composite feature extraction method to capture more feature points that are easy to track,establishing a weight matrix in the context area will give higher weights to areas that are roughly the same as the target's trajectory.To make tracking more accurate,Kalman filter is introduced into the algorithm,figure out where the next target might be..In the case of camera jitter,the ability of the appearance model to describe the target is greatly improved,will improve the accuracy and robustness of video target tracking.To compare the experimental results of multiple groups of improved algorithms with those of the original algorithm.In the Blackboy data set with severe jitter,the original algorithm has a success rate of only 25.5 %,while the improved algorithm in this paper can reach 87.04 %.In order to eliminate the tracking failure caused by strong occlusion,an improved spatiotemporal context tracking algorithm based on sparse representation is proposed in this paper.Combine sparse representation algorithm with space-time context algorithm.Imitate the relationship between the object and context in the original algorithm,build foreground and context dictionaries,using of sparse properties to judge which features were blocked or otherwise disturbed.By shielding these features,the high probability weight is given to the features with good extraction effect,and participate in template update and construct new conditional probability.The improved method can greatly improve the tracking effect in the case of occlusion problem.This paper introduces several data sets to compare with the improved algorithm and the excellent tracking algorithm in the current tracking field,and analyzes the results.In Jogging,where severe occlusion occurs,the original algorithm has a tracking success rate of 19.5 %,and the improved algorithm has a success rate of 94.5 %,which can effectively fight the occlusion problem and stabilize tracking.