Research On Pedestrian Tracking Method For Rotor Flying Robot

Rotary-wing flying robot have the characteristics of small size,high flying flexibility,and have a wider application prospect than other types of robots.Aiming at the lack of visual perception of rotary-wing flying robots,combined with the target tracking technology in computer vision,the pedestrian tracking method based on the rotating-wing flying robot is studied.The main content of the thesis can be summarized as follows:1.A pedestrian tracking algorithm based on adaptive spatial regularization is proposed.Aiming at the boundary effect problem in related filtering tracking methods,the adaptive spatial regularization term is introduced into the classic filter model,and the regular weight is established in adjacent frames.Therefore,the regularization weight of the current frame can be adjusted adaptively,which can effectively reduce the influence of the boundary effect.The algorithm uses an adaptive aspect ratio scale estimation strategy in the process of scale estimation to improve the accuracy of scale estimation in complex scenes.At the same time,it adopts a model update strategy based on color histogram similarity discrimination to suppress model drift and speed up tracking.Through experimental evaluation on multiple data sets,it is verified that the algorithm has good tracking performance,better than other compared algorithms in tracking accuracy and success rate,and has strong robustness in complex task scenarios.2.A pedestrian tracking algorithm DuSiamRPN with dynamic search area and template update is proposed.This algorithm designs a dynamic search area selection strategy to dynamically adjust the size of the search area during tracking according to the characteristics of the high-altitude angle of view of the rotary-wing flying robot.And use the network structure of cropping-inside residual module to eliminate the influence of the padding in the deep network.In addition,a template update network is designed in the algorithm to help the algorithm adapt to various appearance changes of the target.Comparative experiments on multiple data sets show that the algorithm has good accuracy and robustness,has good resistance to various interferences such as occlusion,and can maintain a high level in small targets and long-term tracking scenarios.3.Aiming at the actual pedestrian tracking task,the thesis designs a complete set of pedestrian tracking system for rotary-wing flying robot,including hardware structure and software flow design.On this basis,YOLOv5 is used as the initial pedestrian detection algorithm,and the two algorithms designed in the thesis are respectively used for pedestrian tracking tasks.The performance characteristics of the two algorithms are analyzed through experimental results,and the effectiveness of the pedestrian tracking method proposed in the thesis is verified.