Improved Shadow Removal for Unstructured Road Detection

One of the greatest challenges for vision-based road detection is the presence of shadows and other vehicles. It’s particularly challenging to detect unstructured road when it has both shadowed and non shadowed area since the presence of shadows can cause hindrance and shape distortion of objects which may result in false detection of road. Shadows can also cause a significant problem in road detection since shadow boundaries may be incorrectly recognized or simply hinder the road detection process leading to a higher false rate detection. To tackle those issues, this paper introduces an effective road recognition system using an image processing method to eliminate or reduce considerably the presence of strong shadows for unstructured road detection. Our method’s main novelties are the use of a simple and effective shadow detection and removal algorithm using bilateral filter combined with a model-based classifier. Shadows are detected using normalized difference index and subsequent thresholding based on Otsu’s thresholding method. After the image-preprocessing step used for shadow removal, illumination invariant of road is estimated and a road probability map is calculated to determine whether or not each pixel belongs to road surface. Extensive experiments are carried out and the results show that our method effectively detect unstructured road areas while being robust to strong shadows and illumination variations. It’s also important to note that the proposed algorithm does not depend on temporal restrictions and is invariant to road shape.