Infrared image and radar data-based night unmanned parking lot scene depth estimation method

Abstract

The present invention provides an infrared image and radar data-based night unmanned parking lot scene depth estimation method. According to the method, firstly, a night vision image data set is established, wherein the night vision image data set comprises original sample images and radar data obtained through pre-classifying the original sample images. The original sample images and the radar data are written into corresponding text files. Secondly, a depth convolution / reverse convolution neural network is constructed and trained by utilizing the night vision image data set. Thirdly, a to-be-processed image is acquired in real time, and the to-be-processed image is input into the depth convolution / reverse convolution neural network. Through the depth convolution neural network, a feature map is obtained. The feature map is input into the reverse convolution neural network, and then the category of each pixel point in the feature map can be obtained. After that, a probability map is outputted. Finally, the probability map is subjected to anti-log transformation to obtain the estimated depth of each pixel point. The test proves that the method provided by the invention can effectively estimate the depth of a night scene. Meanwhile, the estimation correctness and the estimation real-time performance are ensured.

Description

technical field [0001] The invention relates to a method for estimating the scene depth of an unmanned vehicle at night based on infrared images and radar data. The method can estimate the spatial position information of the scene in the infrared image. Background technique [0002] Image depth estimation refers to obtaining depth and distance information from images, which is essentially a depth perception problem. Restoring the 3D depth information of a scene from one or more images of the scene is a basic research topic in the field of machine vision, and has important applications in the fields of robot motion control, scene understanding, and scene reconstruction. [0003] Depth estimation techniques mainly include binocular depth cues and image sequence-based depth estimation methods, both of which rely on feature differences between images. For monocular image depth estimation, the classic method in early research is "shape from shading". This algorithm is based on s...

AI Technical Summary

Problems solved by technology

This method first obtains a series of small areas in the infrared image with similar texture and brightness through superpixel segmentation, that is, superpixels; then trains the PP-MRF model to establish the nonlinear relationship between the superpixel panel parameters and their corresponding depths, so as to realize Estimating the given superpixel depth information, the disadvantage of this method is that it needs to manually select features and cannot mine deep feature information of the image.

Images