Vision-based parking lot drivable area detection and local map construction method

Abstract

The invention provides a vision-based parking lot drivable area detection and local map construction method. The method comprises the following steps of S1, collecting a parking lot image data set; s2, marking an actuatable area of the image data set at a pixel level; s3, training the data set by using a semantic segmentation network to obtain optimal model parameters; s4, inputting a new video stream to obtain a detection result of the drivable area; s5, back-projecting a detection result to the ground to obtain an actuatable boundary point cloud; s6, simulating the operable boundary point cloud B into laser data; and S7, constructing a local grid map in combination with a speedometer and a gmapping algorithm with a sliding window. According to the method, the boundary of the drivable area is trained by the neural network, the boundary is simulated into the laser point cloud, and the grid map is constructed by using the sliding window gmapping algorithm, so that the grid map is constructed by only depending on the monocular camera under the condition that the ground is flat.

Description

technical field [0001] The invention relates to the field of robot mapping, and more specifically, to a vision-based method for detecting a drivable area of ​​a parking lot and constructing a local map. Background technique [0002] With the rapid development of science and technology, smart travel has become more and more people's way of life. Traditional cars do not have the characteristics of intelligence, and need to manually complete operations such as overtaking, changing lanes, and avoiding obstacles. High, but also greatly reduces travel efficiency and ride comfort. The latest smart cars such as Tesla Model 3 can perform operations such as autonomous lane changing, remote calling, and automatic parking. As an L3 smart car, Audi A8 can realize functions such as autonomous overtaking on complex road sections and autonomous driving on congested road sections. Although smart cars have achieved partial automatic driving functions, they all rely on expensive sensors, such...

AI Technical Summary

Problems solved by technology

[0003] The environment of the underground parking lot has its particularity, the light is stable, and will not affect the normal operation of the visual sensor; the ground undulations are not large in a small area, which makes it possible to establish a local grid map purely visually; the GPS signal of the underground parking lot is weak , unable to perform accurate positioning; most smart cars use three-dimensional laser radar as a sensor for static object detection and positioning and mapping, ultrasonic or millimeter wave sensors for dynamic object detection, and visual sensors for road recognition

Since 3D lidar is too expensive, the production cost cannot be reduced in a short period of time, which greatly limits the popularization of smart cars

Images