Research On Pedestrian Detection And Tracking Method Based On Rotor Flying Robot

The in-depth study of deep learning has brought machine vision into a new development platform.Combining the relevant results of the two and applying them to real life can provide new ideas for the development of human activities.The video sequences taken from the angle of the UAV face more technical difficulties in the detection and tracking,and the existing detection and tracking algorithms focus on the detection and tracking performance of a single angle,so the existing algorithms need to be improved Adapt to the currently used video image sequence environment.In view of the above problems,this article will discuss the three aspects of the construction of UAV platforms,the improvement of pedestrian detection algorithms,and the optimization of pedestrian tracking algorithms.First of all,according to the mission requirements,the drone platform was built independently.In order to make the UAV more able to track specific target flight and collect video information according to the instructions,this paper builds a control system,visual positioning system,tracking system to enable the UAV to fly to a corresponding location with a specific pedestrian and reach the destination area for pedestrian video Image Acquisition.Secondly,this paper studies the pedestrian detection algorithm based on Faster R-CNN as the basic framework.The detection algorithm is designed for the image features acquired by the UAV.First,based on Faster R-CNN's good detection accuracy for small targets,the Faster R-CNN algorithm is determined as the basic detection framework of this paper.At the same time,according to the characteristics of the data,the clustering method is used to make statistics on the shape of pedestrians,so as to improve the original anchor structure in the algorithm.Regarding the pedestrian occlusion problem,this paper proposes a new regression loss function to improve the detection accuracy of pedestrians in crowded and occlusion situations by balancing the positions between candidate boxes.Finally,this paper studies the pedestrian tracking algorithm based on UAV.First of all,according to the characteristics of the image taken by the drone,the Camshift tracking algorithm with superior calculation amount is selected as the basic algorithm.Because the range of motion of pedestrians is not concentrated and the motion scenes are complex and changeable,this paper proposes a target model based on multi-feature fusion to strengthen the description of pedestrians.At the same time,due to the different ability of different features to describe different objects,this paper designs a weighting coefficient fusion algorithm,which can automatically assign corresponding weighting coefficients according to the degree of feature contribution.In order to improve the detection accuracy of pedestrians on the occlusion problem,this paper proposes a strategy to merge Camshift algorithm with EKF,and improve the robustness of the occlusion situation through the EKF prediction mechanism.Finally,set up the occlusion discriminator,aiming to make EKF prediction when necessary and reduce the amount of calculation.Experiments show that the detection accuracy of the detection algorithm can reach 98.5%,and the tracking algorithm can also accurately track the target after the pedestrian appears to be blocked.