Research On Object Tracking Algorithm Of Home Service Robot In Complex Scenes

As a leading technology of machine vision,moving target tracking has always been a hot research object in the field of artificial intelligence,and plays an important role in the field of computer vision.As a comprehensive discipline,it covers many theories and knowledge such as image processing,control engineering,computer hardware and software technology,sensors and mathematics,neural network,artificial intelligence and so on.It has broad application prospects and theoretical value in many fields,such as visual navigation,aerospace,military guidance,intelligent traffic monitoring,family intelligent service robot and so on.Home service robot is an important application area of target tracking technology.Its working environment background is more complex,and the tracking target is complex and diverse and unknown.TLD(Tracking-Learning-Detection)is an excellent single target tracking algorithm which has been developed in recent years,and can be able to achieve long and stable tracking of the target.But in the real complex scenario,there are still some shortcomings in the algorithm,which can improve the robustness and real-time performance of tracking target occlusion.This paper will focus on the analysis and expansion of the shortcomings of the TLD algorithm in complex scenes,and optimize it to improve the tracking performance of the algorithm.The main contents of this paper are as follows:(1)Researching the algorithm flow and frame of TLD.Through the four modules of the TLD algorithm:detection module,tracking module,learning module and integrated module learning.At the same time,the key part of the algorithm,P-N learning process and the median flow tracking algorithm based on LK optical flow method are studied in depth,and the shortcomings and problems of the algorithm are found.(2)In the algorithm tracking module,the target tracking point selected by the tracker random sampling can not effectively represent the target feature problem.A scheme based on key feature points detection and tracking is proposed,which ensures that the tracking points of samples can be tracked reliably and improve the robustness of algorithm to target occlusion.(3)By introducing Kalman filter into the algorithm detection module,we predict the approximate location of the target in the current frame,and then reduce the detection scope of the scan window,reduce the useless calculation amount of the algorithm,and improve the processing speed of the algorithm.(4)Through the contrast experiment,we verify the optimized algorithm.The optimized algorithm overcomes the shortcomings of the original algorithm,and improves the robustness of the algorithm to target occlusion and the speed of algorithm.