Research On Moving Object Detection And Tracking System Based On Mobile Platform

With the development of technology,mobile robots have been increasingly applied in various fields,especially in vision-based motion object detection and tracking,which is widely used in transportation,robot navigation,human-computer interaction and other fields.However,motion object detection and tracking technology still faces challenges in complex environments,and mobile platforms are prone to losing targets.This article proposes a mobile platform-based motion object detection and tracking system to address the above issues.The design of the upper-level detection and tracking system is emphasized,and extensive research is conducted on motion object detection and tracking algorithms.The system is capable of re-detecting and tracking targets and can be implemented on a mobile platform.The feasibility of the algorithm is tested in a ROS simulation environment.Based on mobile platform experiments,it is demonstrated that the algorithm has good stability and accuracy in detecting and tracking moving targets.The research in this article mainly includes the following four parts:(1)To achieve detection and tracking of moving targets on a mobile platform,the upper-level detection and tracking system is designed,including hardware and software systems.The Turtlebot2 equipped with Kinect v2 is chosen as the hardware system due to the significant performance improvement of Kinect v2 over Kinect v1.The ROS operating system is selected as the development platform,and the software system architecture and implementation process for the mobile platform are designed based on ROS.The construction of the software and hardware lays the foundation for detecting and tracking moving targets on the mobile platform.(2)To achieve motion object detection on the mobile platform and improve the accuracy and real-time performance of small target detection,deep learning-based object detection algorithms are studied and transplanted to the ROS framework for testing.The YOLOv3-Tiny algorithm is selected as the detection algorithm due to its fast speed and ability to be ported to low-performance devices.The multi-scale fusion structure of the YOLOv3-Tiny algorithm is improved,and a human head-shoulder dataset is created and processed using K-means clustering to optimize the model’s accuracy in detecting small targets.The final test based on the ROS framework achieves a detection speed of 162.3 FPS.(3)Traditional tracking algorithms,including TLD(Tracking Learning Detection),MIL(Multiple Instance Learning),and KCF(Kernel Correlation Filter),are studied to address tracking instability and target box drift issues.Using center pixel error,pixel error standard deviation,and FPS as indicators,the tracking performance of the three algorithms is evaluated in different complex environments.To address the problem of tracking failure caused by changes in target scale and occlusion,a re-detection and tracking method based on the optimized YOLOv3-Tiny model is proposed.The optimized YOLOv3-Tiny model is combined with the three tracking algorithms to achieve re-detection and tracking after tracking failure.The evaluation of the three algorithms combined with the optimized YOLOv3-Tiny model in actual videos shows that the optimized YOLOv3-Tiny＿KCF algorithm has an average pixel error of 14.67 pixels and an average FPS of 57.26,demonstrating higher tracking accuracy and stability.(4)The implementation of the mobile platform detection and tracking system is studied,and the algorithm is tested based on both the ROS simulation environment and Turtlebot2.First,a mobile platform model and simulation environment are built in the ROS simulation environment,and the feasibility of the algorithm is verified using two scenarios: target rotation and target straight motion.The algorithm is then tested using the Kinect v2 depth camera and on the Turtlebot2 to achieve autonomous detection and tracking of moving targets on the mobile platform.