Research On Trajectory Recovery Of Indoor Dynamic Object(Pedestrian) Based On Binocular Vision

In recent years,with the development of the computer vision technology,the robot,the unmanned aerial vehicle,the unmanned aerial vehicle,the vr/ ar and the like are widely used in all aspects of the life,and the key technology used is the visual slams technology,the machine carrier with the vision sensor can be made to complete its own positioning in an unfamiliar environment and build an incremental map of the surrounding environment in accordance with its own location,but as a large number of dynamic objects(mainly moving pedestrians)are present in the indoor environment,so that the robot can generate a large error in the positioning of the robot,can not complete the high-precision positioning of the robot,and is a dynamic obstacle in the travel process of the robot from the starting point to the end point,the greatest difficulty of the use of the robot in the room is that there is a large number of moving objects in the room that have a negative effect on the path planning itself,in which case the technique of how to recover the moving track of the indoor moving object is studied,and a robot obstacle avoidance algorithm is designed aiming at the dynamic obstacle formed by the indoor moving object,the main research contents of the invention are as follows:(1)Aiming at the problem of the defect that the pedestrian recognition results in the original model based on yolo network include too many backgrounds,the pedestrian image data is divided into "head","body" and "leg" training to generate a new pedestrian detection model,and to detect pedestrians in the image,the detection results are based on the improved mask input matrix based on the grabcut image cutting algorithm.Finally,the pedestrian image without external environment is extracted.By using the new training pedestrian detection model and the improved image cutting algorithm,the pedestrian detection results are reduced by 39.7% compared with the previous ones,and the pedestrian extraction results are more accurate.(2)Aiming at the problem of calculating the position coordinates of pedestrians in binocular cameras,the position of pedestrians in each frame is calculated by means of triangulation principle.In view of the fact that the triangulation principle is too dependent on extracting the feature points of the pedestrian body,matching the feature points of the same name correctly and calculating slowly(1.5 frames / s),the algorithm has the defect of accurately calculating the pedestrian position according to the triangulation principle.A fast pedestrian location method based on pedestrian detection is proposed,which is based on the rectangle box information(rectangle width,height and center pixel coordinates)detected under the yolo network.The relative error of pedestrian location is better than 5%,and the calculation time at PC can reach 9.8 frame / s,and the mathematical model of pedestrian coordinates and rectangular frame information of pedestrians is established,and the relative error of pedestrians location is better than 5% at the end of the PC.Meet the real-time and high-precision requirements of pedestrian trajectory reconstruction.(3)Aiming at the problem of restoring human motion track in video,based on the results of pedestrian detection based on yolo network,the gray histogram features of the images containing rectangular frames of pedestrians are taken as the features of the same pedestrian set in different positions.Constantly update the features of the pedestrian set,when entering a new pedestrian image to be detected,extract its gray histogram feature matching to represent the pedestrian features in the pedestrian set,match the added pedestrian set,and build a new pedestrian set if it has not been matched.Finally,the adjacent coordinates between the front and back of the pedestrian collection are connected to get the trajectory of the pedestrians.(4)Aiming at the problem that the robot can successfully avoid the dynamic pedestrian in binocular vision during the process of moving from the beginning to the end,the moving area of the robot is divided into static and dynamic regions on the basis of restoring the trajectory of indoor mobile pedestrians.The adaptive robot path planning algorithm is adopted.When the robot moves to a static region,the global path planning algorithm based on A* algorithm is adopted through the region,and when the robot moves to a dynamic region,A local path planning algorithm based on artificial potential field is adopted.