Research On Dynamic Object Following Of Healthy Caring Robot Based On Monocular Vision

At the end of 2018,the population aged 60 and above has reached 241 million in China,accounting for 17.3% of the total population,and the trend of population aging is still accelerating.In view of the current social aging of the population and the scarcity of human resources,and actively responding to the national 13 th Five-Year Plan,the development plan of the robot industry,the wisdom for the elderly,and other policy calls,this paper takes robot hardware devices as the carrier,focuses on the dynamic object following technology with object detection and object tracking as the core.It develops the key technologies and prototype system for intelligent healthy caring robots,to provide high-quality healthy caring services for the elderly.While ensuring the quality of life of the elderly in their later years,it will better liberate the social labor force.This paper proposes a multiple object tracking algorithm based on multi-feature fusion optimal correlation based on object detection.The tracking accuracy of dynamic objects in multi-person situation severely restricts the follow-up performance of healthy caring robots.Especially in moving scenarios,the non-rigid deformation of people and the occlusion overlap of the human interaction,the camera and the object two-way motion are easy to cause the exchange of personnel ID,which make the failure of robot's following.In order to solve the above problems,this paper uses filter tracker,person re-identification network and scene depth prediction network to construct one multi-feature model of motion information,appearance information and depth value information,and through the combination of stratification strategy and HM&KM algorithm,to solves the optimal association matching,and achieves multi-object tracking.Compared with the existing methods on the MOT16 dataset,our method can effectively reduces the number of ID exchanges with the similar tracking accuracy.Focusing on the healthy caring scene,this paper builds a dynamic object following framework based on the multiple object tracking algorithm.Firstly,based on Faster RCNN detection algorithm and VGG16 network,this paper applies Soft-NMS mechanism and data augmentation technology to optimize and train the detection model,which improves the accuracy precision of object detection in the healthy caring scene.Then according to the camera configuration and placement position,this paper applies the camera model and Zhang Dingyou the calibration method to construct the monocular vision measurement model,achieving the measurement of the object angle and distance.Finally,by combining the algorithms of detection,tracking and measurement,and outputting the measurement result of the fixed ID,the construction of dynamic object following framework is completed.Based on the framework of dynamic object tracking,this paper uses PadBot U1 robot and Turtlebot2 robot as carriers,further integrates speech recognition and synthesis technology and remote video technology,designs and implements the function modules of dynamic object real-time follow-up and call for help or care under abnormal situation,gradually builds a complete follow-up caring system for healthy robots.Through technical performance test and system function test,it is proved that the system has strong robustness and platform portability.