A Supervised Learning-Based Deep Autoencoder for Road Segmentation

Abstract

A road segmentation method based on a deep autoencoder based on supervised learning. Aiming at the low segmentation accuracy and poor real-time performance of the traditional segmentation method, the semantic segmentation method based on the full convolutional network has a complex network structure and cumbersome parameter tuning. , long training period and other issues, a supervisory layer is added to the traditional autoencoder model, and features that are beneficial to road image segmentation are extracted through a supervised learning mechanism to achieve semantic segmentation of road images. The supervised learning mechanism of the present invention promotes the network structure to focus on learning information such as contours and boundaries of regions while ignoring image details irrelevant to segmentation, thereby achieving better road segmentation effects. Moreover, the method model proposed by the present invention is simple, and the training time and running time are far lower than the Segnet network, which is extremely critical for road recognition with high real-time requirements.

Description

technical field [0001] The invention belongs to the technical field of computer vision, relates to intelligent vehicles and unmanned vehicles, and specifically relates to a road segmentation method based on a supervised learning deep autoencoder. Background technique [0002] The road environment perception of unmanned vehicles has always been a research hotspot, and the road environment perception method based on machine vision is one of the research focuses. The method realizes road image segmentation, and image semantic segmentation means that the machine automatically segments the object area from the image, recognizes the content in it, and determines the drivable area of ​​the unmanned vehicle. [0003] Most of the traditional image segmentation methods are based on the feature extraction of the image itself. It is necessary to generate different regions on the image first, then extract features on the region, and classify and merge the regions to obtain the final sema...

AI Technical Summary

Problems solved by technology

[0004] The problem to be solved by the present invention is: the traditional segmentation method has low segmentation accuracy and poor real-time performance, and the semantic segmentation method based on the full convolutional network has complex network structure, cumbersome parameter tuning, and long training period. A Simple and Effective Road Segmentation Method is Proposed

Images