Object-level rapid scene recognition method based on semantic point cloud

Abstract

The invention discloses an object-level rapid scene recognition method based on semantic point clouds, which is characterized by comprising the following steps of: determining a main object before mapping: processing three-dimensional point clouds collected by a sensor through a semantic segmentation network, and determining the main object according to the distribution of the semantic point clouds; constructing a k-dimensional tree structure while mapping, establishing a polar coordinate system by taking the main object as a pole and taking the connecting line direction of the sensor and the main object as a polar axis, and constructing a descriptor; during scene recognition, calculating the similarity between a historical frame and a current frame: searching descriptors corresponding to nodes similar to the current column vector obtained by averaging according to the row in a k-dimensional tree structure constructed by taking the column vector obtained by averaging according to the row as the nodes, and taking the descriptors as candidates; and calculating the accurate pose of the current frame: calculating the accurate pose of the current frame based on the observation positions of the main object in the current frame and the matching frame. According to the method, the descriptor is established by taking the main object as the center, and the descriptor independent of the observation position can solve the problem that scene recognition cannot be carried out when the positions of two frames are not overlapped.

Description

technical field [0001] The invention belongs to the fields of robots and automatic driving, and in particular relates to an object-level rapid scene recognition method based on semantic point clouds. Background technique [0002] SLAM synchronous positioning and map construction technology has a wide range of applications in the fields of autonomous driving, drones, robots, VR and AR. SLAM specifically refers to: the device starts to move from an unknown position in an unknown environment, locates itself according to the data collected by the sensor and the environmental map constructed during the movement process, and builds an incremental map based on its own positioning the process of. [0003] In the actual SLAM process, the scene recognition technology in the SLAM system is required to realize loop detection and relocation. This scene recognition technology needs two frames of sensor data before and after, and can identify whether the front and back are the same scene...

AI Technical Summary

Problems solved by technology

[0006] Aiming at the deficiencies of the prior art, the present invention proposes an object-level rapid scene recognition method based on semantic point cloud, aiming to solve the problem of the existing loopback detection algorithm and relocation algorithm for the current and historical moments of the vehicle in the same scene. When the frame positions do not coincide, the system cannot detect loopbacks and cannot relocate

Images