Semantic point cloud generation method and device based on laser radar and visual fusion

Abstract

The invention relates to a semantic point cloud generation method, device and equipment based on laser radar and visual fusion and a computer readable medium. The method comprises the following stepsthat a point cloud cluster map is obtained based on laser radar point cloud data; an object detection classification map is obtained based on image data of a vision sensor; according to position information of the detected object and classification tag probability information in the object detection classification map and position information of laser clusters in the point cloud cluster map, classification label probabilities, matched with the position of the detected object, of the laser clusters are obtained; and according to the classification label probability of the laser clusters, classification label of the detected object is marked onto the laser clusters, and semantic point cloud data are generated. The classification label probabilities of the laser clusters at corresponding positions are calculated based on the classification label probability of the detected object, the laser clusters are subjected to classification label marking, and therefore accurate point cloud laser cluster classification label marking is achieved, and accuracy of semantic annotation of the generated semantic point cloud is improved.

Description

Technical field [0001] The invention relates to the technical field of object detection, in particular to a method, device, equipment and computer-readable storage medium for generating a semantic point cloud based on lidar and vision fusion. Background technique [0002] Lidar is a radar system that emits a laser beam to detect the characteristic quantity of a target object. When using lidar detection equipment, such as a mobile robot loaded with lidar, to perform target detection such as target positioning, lidar has the defects that the returned point cloud data is sparse and the amount of information is small, resulting in poor detection results of lidar in a dynamic environment. Good, the accuracy is not high. [0003] In order to solve the problem of inaccurate detection results caused by sparse lidar point cloud data, a target object detection technology that combines lidar and visual detection has emerged. How to fuse lidar point cloud data and visual detection image data ...

AI Technical Summary

Problems solved by technology

However, due to the above-mentioned existing methods, due to the position difference between lidar and visual detection, and the object detection frame of visual detection contains a lot of background information, an object detection frame may correspond to multiple point cloud laser clusters in positional relationship, and it is often difficult to identify the object The detected classification labels are accurately marked into the corresponding point cloud laser clusters, resulting in large errors, resulting in inaccurate semantic marking of the generated semantic point cloud data

Images