Research And Implementation Of Recognition And Tracking Algorithm Based On Visual Moving Target

The coordination of heterogeneous robots is an important research direction to expand the scope of the action of the single robot and improve the success rate of the robot to complete the complex tasks.The air-ground cooperative system is a typical heterogeneous robot system.In this paper,the target recognition and tracking of UAV Based on the ground to ground robot cooperative system is studied.On the basis of theoretical research of computer vision technology,UAV recognition and tracking of ground robots are completed.This paper expounds the heterogeneous robot cooperative system,analyzes the requirement of the system,and designs the target tracking scheme based on the computer vision.This scheme designs the hardware implementation platform and the work flow of the target tracking.The computer vision algorithm theory is the key content of the paper.The image preprocessing in the space robot cooperative system mainly works on the image restoration.Image degradation is mainly derived from natural weather,sensor vibration,random channel noise,and so on.This paper mainly deals with the research of image denoising.First,we use K-means algorithm to classify image blocks,learn dictionary from different sub categories,and reconstruct images at last.In the process of reconstruction,we use non local similarity method to improve the effect of image restoration.After de-noising the image,it is the work of target recognition.Image matching method is adopted,and SIFT features with better robustness are applied.In the matching process,the PCA-KD tree algorithm is used to improve the matching.In order to further improve the matching accuracy,Ransac algorithm is used for the two match.Then the perspective transformation is used to find the position of the target.After getting the target position,it is the work of target tracking.Multi-scale target tracking algorithm based on Kalman filter correction is proposed.The algorithm uses Kalman filter to generate distance weight,uses weighted Bayesianfilter to correct tracking position,and performs multi-scale template acquisition on the determined location to adapt to scale change.When the classifier parameters are updated,the adaptive learning rate is introduced.The tracking flight experiment of the four rotor unmanned aerial vehicle was studied,the camera model was constructed,and the position relationship between the unmanned aerial vehicle and the tracking target was transformed by coordinate.Based on this,the navigation parameters of the UAV are calculated,and then the UAV is finally controlled to follow the UGV by air.The experiment shows the effectiveness and feasibility of the computer vision algorithm in this paper.