Research On Moving Target Tracking Method Based On Machine Vision

Visual target tracking is a key research topic in machine vision,which is widely used in military and civil fields,with important research significance and research value.The KCF target tracking algorithm,TLD target tracking algorithm and deep learning combined with KCF algorithms in visual target tracking algorithms have been studied and improved accordingly.The relevant research contents and conclusions are as follows:1)Aiming at the problem that the single feature of KCF algorithm can not adapt to the complex scene in the tracking process and can not solve the problem of scale variation of the target,a scale adaptive correlation filter tracking algorithm combining FHOG and LBP features is proposed.The FHOG feature and LBP feature are adaptively weighted fused to locate the target.Then the scale pyramid is used to estimate the scale of the target.The simulation results show that the accuracy and success rate of the improved algorithm are improved,and it has better robustness.2)Aiming at the problem that the classical TLD tracking algorithm is affected by occlusion,illumination,interference,rotation and scale variation,which results in low tracking accuracy,high computational complexity and poor real-time performance,an improved TLD tracking method based on HOG-SVM is proposed.The HOG-SVM is combined to replace the 2bit BP feature and integrated classifier in the original TLD algorithm;the KCF tracking algorithm replaces by median optical flow method.Based on the HOG-SVM+KCF tracking algorithm,the sliding window method is improved.The simulation results show that under the circumstance of changing background environment,the tracking accuracy and tracking speed of the improved algorithm are improved,and the real-time performance is enhanced.3)Aiming at the problems of target tracking algorithm tracking in complex background,such problems as feature redundancy,low signal-to-noise ratio,and scale variation,a multi-scale target robust tracking algorithm based on convolutional neural networks is proposed.CNN features on each layer of weak trackers are assigned different weights,and a strong tracker is combined by weighting to evaluate the position of the target in the current frame,and the weight parameters of the weak tracker are updated online.Then the scale filter is trained to estimate the scale of the target.The simulation results show that the algorithm has good accuracy and robustness,and better tracking of target scale variation.Finally,the simulation experiments of complex background video series under intelligent monitoring are carried out to verify the effectiveness of the three algorithms.Figure47;Table9;Reference 43...