Research On Pedestrian Video Detection And Tracking Technology Based On Mobile Platform

The position of target detection and tracking technology in the field of computer vision should not be underestimated.With the development of social production and life and science and technology,the research on video pedestrian detection and tracking technology based on mobile platform also needs to be put on the agenda.It can be applied to traffic video monitoring,analysis of athletes' competition,human-computer interaction,and design of tracking system.Therefore,this paper focuses on the video recorded by the mobile robot in the designated range and studies the rapid and effective detection and tracking of pedestrians in video surveillance.The main work and research results of this paper include the following aspects:Build mobile robot platform based on ROS.Through the learning of ROS system,build the required platform for EAI Dashgo mobile platform and configure the corresponding environment.Through the combination of laser radar and ultrasonic detector,SLAM mapping is completed,and the combination of global path planning algorithm and teb local path planning algorithm is used to complete single-point multiple times and multi-point multiple times navigation within a short distance,finally realizing the patrol work within the specified range.At the same time,visual sensors record and save video for later use.Collect and make the image data set of pedestrian detection.According to the research requirements of this paper,a total of 16,420 untransformed pedestrian detection data sets 2019 were collected.Select the network model suitable for pedestrian detection in this paper.Through theoretical and experimental analysis and comparison,YOLOv3 is selected as the network model for pedestrian detection,which achieves a good balance between detection accuracy,model parameters and calculation time.At the same time,the influence of model parameters on the detection results was compared and analyzed,and relevant experiments and analysis were carried out by using the pedestrian detection data set 2019 produced in this paper.Through tests,the detection accuracy was improved by 5.0%,which verified the high efficiency of the dedicated pedestrian detection YOLOv3 network model in pedestrian detection.Select an appropriate pedestrian tracking network model.This paper introduces the mainstream target tracking technology,selects Deep_Sort as the pedestrian tracking technology in this paper through comparison,and makes a test and comparison experiment on MOT data set.The high cost performance of Deep_Sort in target tracking was verified.Pedestrian detection and tracking system based on YOLOv3 combined with Deep_Sort.The video was recorded and filmed by the mobile robot for indoor inspection,and the obtained video was tested based on YOLOv3 combined with the pedestrian detection and tracking system of Deep_Sort,which verified the feasibility of combining the two algorithms to detect and track pedestrians in the video.To sum up,in this paper,the mobile robot based on Gmapping mapping method to map the unknown environment,based on A*and Dijistra algorithm to complete the safety inspection in the known environment.During the inspection process,the recording and recording of the video are collected.The modified pedestrian detection network detects pedestrian targets in the video.The Deep_Sort target tracking algorithm establishes tracking of the detected targets,and the effect is better.