Key Technologies Of "Unmanned Supermarket" Autonomous Mobile Robot Based On 3D Vision

In recent years,“new retail” has been warmly welcomed by people because of its convenience.As one of the development trends of “new retail”,“self-service supermarket” received great attention.However,due to the lack of anti-theft measures,the identification technology is not mature and so on,it can not be promoted massively.Research shows that the real "unmanned supermarket" without human shop assistants and customers is an important development direction of the supermarket industry in the future.In order to realize the development from “self-service supermarket” to “unmanned supermarket”,technologies of vision-based mapping,navigation and commodity detection of supermarket based mobile robots are studied.The specific work is as follows:First,a two-dimensional grid-map construction method based on ORB_SLAM2 algorithm is proposed to solve the problem that the existing visual SLAM algorithm can not directly construct the grid-map for subsequent navigation.Fused by the QR code information,the position and shape of the shelf in the grid map are corrected and semantically assigned.The experimental results show that the semantic map constructed by this method has a shelf displacement error of 12.41 cm and a tilt error of 1.68°,which can be used for subsequent navigation work.Second,the target commodity detection method based on YOLOv3 algorithm is realized.The experimental results show that the product detection model can effectively detect 6 types of beverage commodity.When the IOU threshold is 0.5,the average detection accuracy rate for each type of commodity reaches 0.92,and the recall rate reaches 0.90,which can basically meet Detect the demand for shelf goods.Third,an improved A* algorithm with adding the cost of the moving direction is proposed.The simulation experiment verifies that the improved algorithm is not worse than the A* algorithm in terms of search space and path length while the number of turns of the path is greatly reduced.In addition,the TEB algorithm is studied,and the best key-parameters of it are obtained through large number of debugging on the mobile robot platform developed by the research group,and the effectiveness of the TEB algorithm under the platform is verified.Finally,based on the above work,a ROS-based mobile robot navigation system in supermarket is designed and implemented.Robots in the system navigate automatically according to the input target commodity name.The test results in the simulated supermarket environment confirm the effectiveness of the navigation method and are expected to be applied to real supermarkets.