Research On Road Detection Algorithms Under Different Light Conditions

With the development of computer vision technology,more and more outdoor scenes of mobile intelligent machine equipment have emerged,from the earlier outdoor intelligent mobile robots and the smart driving vehicles that have recently been pushed to the research outlet.For visual navigation,accurate detection of the road section is particularly important.However,in actual situations,roads vary widely and the environment is different,so research in this area actually has enormous challenges.Based on the above background,this paper proposes a new algorithm for road edge detection,which is mainly for road images collected under different weather conditions,including roads with no shadows during the day and night,roads with shadows under the sun and roads with shadows under the street lights at night.The paper mainly consists of three partial modules.In the image acquisition and feature extraction module,this paper proposes an feature extraction method based on RGB,HLS and HSV three color space.The system firstly separates the image of day or night by using the support vector machine classifier,and combines the features of the shadowed and unshaded road images in the three spaces to make a shadow image detection judgment.For the case of the shadowed and unshaded road images,the system adaptively selects the corresponding feature space component,so that the pixels of the shadowed area and the unshaded area in the image are unified or their gap is reduced.The gray value of the selected color space is stable and provides a basis for subsequent segmentation algorithms.The image segmentation and denoising module is after the extracted color space.This paper proposes an improved segmentation algorithm based on the Otsu.By using the median to replace the mean,when inter-class variance and the weighted intra-class average variance ratio is max,the threshold is chosen to split the image.While reducing the complexity of the algorithm,compared with the traditional algorithm segmentation,it is found that the improved algorithm is more suitable for the images collected in this paper.In order to remove the noise,in addition to the morphological method,the improved noise removal algorithm based on the spatial filtering idea filter is used to remove the noise of the road part.In the line fitting module at the edge of the road,we extract the point set by column,and combine the least square method,Hough and RANSACalgorithm to perform the wild value elimination and straight line fitting,thereby improving the accuracy of edge detection.Finally,the fitted line graph and the original image are merged according to the weight in the same channel,so that the line of the road edge detected by the algorithm and the actual road edge can be visually compared.In this paper,the research on road detection under different luminosity has achieved certain results.According to the experimental results,the average accuracy of road detection can reach 85%,and the real-time performance can reach 42ms/frame.To some extent,it solves the problem of road detection in the case of roads with no shadows during the day and night,roads with shadows under the sun and roads with shadows under the street lights at night.