Method and device for detecting pole lift applicable to self-driving vehicles

Abstract

Embodiments of the present disclosure disclose a method and an apparatus for detecting a pole lift that can be used in an autonomous vehicle. A specific implementation of the method includes: obtaining the area where the pole lift at the exit or entrance is located according to the map; obtaining point cloud data of a predetermined number of consecutive frames in the area during the movement of the vehicle, and for each frame of point cloud data, based on the frame point The cloud data detects the existence probability and angle of the mast; judge the state of the mast according to the existence probability and angle of the mast judged in the point cloud data of a predetermined number of consecutive frames; if the state meets the preset traffic conditions, the output is allowed current information. This embodiment uses a geometric method to fit the posture of the lifting rod, and outputs the state of the lifting rod through a sequential state machine, which can accurately detect the position and state of the lifting rod, and improve the safety of the self-driving vehicle.

Description

technical field [0001] Embodiments of the present disclosure relate to the field of computer technology, and in particular to a method and device for detecting a pole lift that can be used in an autonomous vehicle. Background technique [0002] Self-driving vehicles need to perform perception modeling of the surrounding environment during driving. For example, when entering and exiting a parking lot or a hotel, it is necessary to recognize the lifting pole (railing that can be lifted) of the exit or entrance; accurately detect the lifting pole The position and the state of the lifting lever are crucial to the planned control of the autonomous vehicle. [0003] Existing pole lift detection methods are usually based on visual images. Similar to general object detection, the detection of pole lifting is also a method of feature learning + statistical learning. It is necessary to prepare pictures related to lifting poles as training samples in advance, and train feature extract...

AI Technical Summary

Problems solved by technology

[0004] Due to the lack of depth information in the vision, the poles detected on the 2D plane often have certain positioning errors when they are projected onto the 3D plane; the pole lift detection based on visual images is greatly affected by the background, which is strongly dependent on the training data and the detection model. The generalization performance has very high requirements; when detecting the state of lifting the pole, it often needs inclined preset anchor points, which is difficult to integrate into the existing general object detection framework

Images