The AGV car is a mobile robot, an automated logistics equipment with autonomous positioning, navigation and obstacle avoidance functions. It is used to build a flexible logistics system in modern manufacturing workshops. Realizing autonomous positioning of mobile robots is the prerequisite and key to robot navigation and obstacle avoidance. At present, the common positioning methods of AGV trolleys include lidar positioning, GPS positioning and visual positioning. Lidar positioning technology is not suitable for AGV car positioning. Although the positioning accuracy is high, it is expensive. It is mainly achieved by measuring the distance and angle of the surrounding environment; GPS positioning technology is positioning by receiving GPS satellite signals. The positioning error is large and easy to be affected. The influence of buildings and trees is not suitable for the indoor positioning of AGV vehicles; visual positioning technology is very suitable for the indoor positioning of AGV vehicles. It is to estimate the position of AGV vehicles by processing image stream data, with strong scalability, rich information, and low price.

Aiming at the current deficiencies of global positioning system and lidar positioning, technicians provide an automatic guided vehicle positioning system and method. A positioning system for AGV trolleys includes: a deep learning module, which uses deep learning methods to train a global three-dimensional point cloud map of the working environment, obtains map learning models and feature matching criteria, and stores them in a cloud server; global mapping module uses positioning The surveying and mapping algorithm constructs a global three-dimensional point cloud map of the working environment, while using lidar to scan the working environment to obtain laser data; builds a local three-dimensional point cloud map of the working environment according to the pinhole imaging principle, and the local mapping module uses sensors to collect image data of the working environment in real time. Including depth images and color images; the matching positioning module searches for local 3D point cloud map information in the global 3D point cloud map according to the feature matching criteria, and restores the attitude information of the AGV car according to the matching information; real-time display module, used to display the working environment The two-dimensional point cloud map displays the attitude information of each AGV car in the global map in real time. The image data includes color images and depth images. The lidar is installed on the top center of the AGV trolley to scan the working environment to obtain laser data and build a global three-dimensional point cloud map. The sensor is installed in front of the AGV trolley to construct a local 3D point cloud map and collect image data of the working environment. The positioning method of the AGV car is to collect the color image and depth image in the environment at a rate of 30FPS; according to Zhang's calibration method, the camera is calibrated to obtain the camera's internal parameters: the camera's focal length (fx, fy) and the camera's aperture center (CX, Cy); Use the pinhole imaging principle to recover the three-dimensional coordinates of any point in the depth image to obtain a local three-dimensional point cloud image of the environment.

Compared with the prior art, the positioning system of the aforementioned AGV trolley has the following beneficial effects: 1. By constructing a local point cloud map of the environment and matching the global point cloud map, a relatively inexpensive Kinect sensor is used to replace the multi-line lidar to realize the AGV trolley. Autonomous positioning, thereby reducing the cost of AGV trolleys. 2. Use the deep learning method to train the global 3D point cloud image to obtain the feature model and matching criteria, which can not only meet the requirements of the real-time positioning of the AGV trolley, but also improve the positioning accuracy of the trolley.