Distance space reconstruction method based on deep learning and monocular vision

Abstract

The invention discloses a distance space reconstruction method based on deep learning and monocular vision. The distance space reconstruction method includes the steps of calibrating image acquired animage acquisition module, marking the target contour, the type, the target inclination angle and the target distance of the images, combining the labeling data into multi-dimensional vector arrays toserve as sample training labels; taking the sample training labels as the input of a deep neural network, training the deep neural network, reconstructing a spatial structure through the deep neuralnetwork, and estimating the target distance estimation through target recognition, frame regression and spatial structure regression. By means of the powerful nonlinear fitting function of deep learning, the distance space information of all the samples is synthesized, distance estimation is performed globally, and accuracy is greatly improved.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence and automatic driving, and in particular relates to a distance space reconstruction method based on deep learning and monocular vision. Background technique [0002] The automatic driving system is an active safety system that can automatically control the operation of the vehicle, including driving, changing lanes, parking, etc., which improves the driving experience and comfort while ensuring driving safety. At present, the automatic driving system mainly uses sensors installed in the vehicle, such as: millimeter-wave radar, lidar, camera and ultrasonic radar, etc., to detect and recognize the driving environment, including lane lines, surrounding vehicles, pedestrians, obstacles, traffic lights, traffic signs, etc. , Safely and efficiently automatically control vehicle driving while obeying traffic rules. When automatic driving detects obstacles, it needs to avoid, and the dist...

AI Technical Summary

Problems solved by technology

[0004] (1) Vehicle ranging technology based on binocular vision, which requires too high installation accuracy of binocular cameras and is not robust enough

[0005] (2) Vehicle ranging technology based on lidar The advantage of this technology is high precision, but the disadvantage is that the cost is too high. A good lidar price is around 600,000

[0006] (3) Vehicle ranging technology based on monocular vision, which has low precision, inaccurate distance estimation, and great difficulty in practical use

Images