3D target detection method for point cloud screening based on image semantic features

Abstract

The invention provides a 3D target detection method for point cloud screening based on image semantic features. The method comprises the following steps: segmenting image data by a 2D semantic segmentation method to obtain semantic prediction; projecting the generated semantic prediction to an LIDAR point cloud space through a known projection matrix, thereby enabling each point in the point cloudto obtain the semantic category attribute of the corresponding image position; extracting related points of vehicles, pedestrians and riders from the original point cloud and forming a view cone; taking the view cone as the input of a depth 3D target detector, and designing a loss function conforming to the characteristics of the view cone to perform network training. According to the invention,a 3D target detection algorithm for point cloud screening based on image semantic features is designed, so that the 3D detection time is greatly shortened and the calculation requirements are greatlyreduced. Finally, the performance of the method on a reference data set KITTI of 3D target detection shows that the method has good real-time target detection performance.

Description

technical field [0001] The invention relates to 3D target detection, in particular to a 3D target detection algorithm for point cloud screening based on image semantic features, and belongs to the field of pattern recognition. Background technique [0002] Point cloud-based 3D object detection plays an important role in many real-life applications, such as autonomous driving, home robotics, augmented reality, and virtual reality. Compared with traditional image-based object detection methods, LIDAR point cloud provides more accurate depth information that can be used to locate objects and describe object shapes. However, due to factors such as non-uniform 3D space sampling, the effective range of the sensor, and object occlusion and relative position, making it different from traditional images, LIDAR point clouds are more sparse and have large differences in density between parts. To address the above issues, many methods use artificially designed feature extraction method...

AI Technical Summary

Problems solved by technology

However, these methods that use all point clouds as input require a lot of computing resources and cannot achieve real-time detection.

Images