Research On Outdoor Target Recognition And Location Method For Quadruped Robot Based On Multisensor Fusion

Quadruped robots are often used to perform various complex field tasks because of their ability to adapt to rugged terrain.In the practical application of quadruped robots,it is often necessary to identify and locate the target and navigator.Compared with tracked and wheeled robots,quadruped robots will produce more substantial bumps when traveling,resulting in severe shaking of the field of vision.In addition,there are also various complex and changeable scenes and unknown disturbances in the outdoor environment,which bring difficulties and challenges to the target recognition of quadruped robots.In view of the above problems,this paper studies the outdoor target recognition and positioning method of quadruped robot based on multi-sensor fusion.The purpose of this paper is to improve the robustness and fault tolerance of outdoor target recognition and positioning of quadruped robot by using multi-sensor fusion technology,so as to deal with complex scenes such as dramatic light changes,hidden targets similar to the background,high reflectivity material interference,violent shaking of vision,insufficient perception of vision,and occlusion of objects,so as to avoid potential safety hazards caused by recognition errors.The specific research contents and innovations of this paper are as follows:(1)A multi-sensor fusion perception system has been built,including depth cameras,LiDAR,GPU embedded industrial control computers,etc.A software platform has been built.Sensors has been calibrated.This system provides a physical basis for algorithm design and experimental verification.(2)A multi-sensor fusion framework based on adaptive federated filtering is proposed.The framework uses Kalman filter,and can adaptively adjust the information distribution factor according to the illumination conditions.It combines multiple information for fault detection and isolation,realizes multi-sensor data fusion,and improves the robustness and fault tolerance of quadruped robot target recognition.(3)A day and night leader recognition and positioning method suitable for quadruped robots has been built.This method can enable quadruped robots to identify leader in real-time,accurately,and stably in complex outdoor scenes,and complete the following function.This article constructs a visual leader recognition method,including SSD object detection algorithm,KCF tracking algorithm,and ReID personnel recognition algorithm.This article also constructs a LiDAR leader recognition algorithm based on reflection intensity.Finally,a fusion framework based on adaptive federated filtering is adopted to fuse the two algorithms mentioned above.This method has been experimentally validated on SCalf quadruped robots.In complex scenarios such as severe changes in light,interference from highly reflective materials,and severe field of view shaking,quadruped robots can stably recognize and locate leader,demonstrating strong robustness and fault tolerance.(4)A operation target recognition and localization method suitable for quadruped robots has been bulit,which can enable quadruped robots to stably recognize hidden task targets in complex outdoor scenes.This article constructs a dedicated Sucro Field dataset,and the object detection algorithm based on YoloV5 is used to realize the high accuracy recognition of common operation targets such as doors,windows and packets.This study combines camera depth information to achieve three-dimensional positioning of operation targets,designs sensor layout to expand perception field of view,and uses a federated filtering fusion framework to fuse the detection results of multiple cameras.To address the problem of multiple targets in the field of vision,human-machine interaction is used for task target screening.This method has been experimentally validated on SCalf quadruped robot,and the results show that the quadruped robot can stably recognize hidden work targets in situations such as severe shaking of the field of view,object occlusion,and insufficient perception of the field of view,and obtain accurate three-dimensional spatial positions of the work targets.This method has strong robustness and fault tolerance.