Vehicle bounding box generation method, device and storage medium in unmanned scene

Abstract

The invention discloses a method, a device and a storage medium for generating a vehicle bounding box in an unmanned scene. Generating a two-dimensional convex hull according to the point cloud data;The bounding box is constructed according to the lidar scanning direction edge in two-dimensional convex hull. Select one of the constructed bounding boxes as the final desired bounding box. The application of the scheme of the invention can improve the accuracy of the result and the like.

Description

【Technical field】 [0001] The present invention relates to computer application technology, in particular to a vehicle bounding box generation method, device and storage medium in an unmanned driving scene. 【Background technique】 [0002] In order to ensure the safety of unmanned vehicles driving on the road, it is necessary to accurately determine the vehicle bounding box, and then estimate the vehicle orientation, which is of great significance for obstacle collision detection and vehicle control planning. The generation of vehicle bounding boxes in a single frame scene is the basis for continuous frame vehicle tracking. [0003] The bounding box refers to encapsulating complex objects in a simple bounding box, and using simple bounding box shapes to approximate complex geometric shapes. [0004] At present, the widely used bounding boxes mainly include axis aligned bounding box (AABB, Axis Aligned Bounding Box) and oriented bounding box (OBB, Oriented Bounding Box). like...

AI Technical Summary

Problems solved by technology

[0006] However, the premise of accurately generating AABB bounding boxes and OBB bounding boxes is that the shape information of the object is relatively complete. In unmanned driving scenarios, due to sensor characteristics and occlusion and Distance factors, etc., the shape information of objects such as vehicles is not complete. Therefore, using existing methods to generate vehicle bounding boxes, the effect is not ideal and the accuracy is poor.

Images