The Research On Road Detection Algorithms Based On Monocular Vision

The key for vision-based road detection is the ability to classify image pixels as belonging to the road surface or not.Vision-based road detection is very challenging since the road is in an outdoor scenario imaged from a moving mobile platform,including shadows,heavy traffic flow and continuously changing background.Monocular vision has low cost,whereas visual information is difficult to be processed,thus road detection algorithm based on monocular vision has become a research hotspot.In this paper,for the specific application of unmanned ground vehicles,we study how to detect road based on monocular vision in noisy road environment and the proposed algorithm for road detection provides a high accuracy and robustness to shadows.The main work and research results are as follows:In order to accelerate the speed of road detection and reduce the interference of useless information,a method of setting the region of interest(ROI)based on the speed of the unmanned vehicle is proposed.ROI is a part of the whole image which is acquired using a calibrated camera.The size of ROI is decided by the speed of the unmanned vehicle.The method of setting the ROI is simple and efficient.In order to be robust to shadows,an algorithm of shadow detection and removal is improved.It makes the algorithm satisfy the real-time requirement.First of all,to detect shadows,color and texture histograms are used to train the shadow detection classifier because shadowed regions tend to be dark,with little texture.Then,using a matting technique estimate a fractional soft shadow value.Finally,estimate the ratio between direct light and environmental light and generate the recovered image by relighting each pixel with both direct light and environmental light.A novel algorithm for road detection based on D-S evidence theory is proposed to improve the detection accuracy.The algorithm gets a road probability map associated with road geometry during the training of the road geometry classifier,and establishes a road probability map based on color features through extracting color information.Then,the algorithm obtains the final road region by combining the previous two road probability maps based on Dempster's combination rule.Finally,the algorithm shows a new way to represent road region as a polygon for the specific application of unmanned ground vehicles.Implement shadow detection algorithm and road detection algorithm based on D-S evidence theory within ROI.Experimental results demonstrate that it is effective at detecting road region in challenging conditions.Its efficiency has been obviously raised,with the same processing speed,when compared to state-of-the-art methods.