Research On The Method Of Identifying The Vehicle's Drivable Area Based On Image Semantic Segmentation

Environment perception for intelligent driving is the key to vehicle driving decision and control,and drivable area recognition is one of the important issues in the field.On the one hand,as the driving environment is complicated and changing continuously,to achieve breakthroughs in such area becomes a daunting task.On the other hand,intelligent driving,by its nature,demands more precision and timeliness for the algorithm,which puts more critical requirements for the real applications.In this paper,the drivable area of vehicle is taken as research object,by optimizing deep learning algorithm,and useing the technology of binocular stereo vision to obtain spatial information of driving scene in 3D,we propose a novel method of semantic segmentation based on convolutional neutral network(CNN),in order to achieve precise and efficient recognition of drivable area for intelligent driving.The main work of the paper is as follows:(1)The key for semantic segmentation of image depends on the image features.In view of the superior feature extraction performance of convolutional neural networks,we construct a semantic segmentation model based on CNN “Encoder-Decoder” architecture to extract the features of drivable area.Considering the real-time requirements,we adopt a shallow network process to reduce the network complexity.At the same time,in order to ensure the accuracy of model recognition,we propose Ournet model by updating and optimizing the “width of network”,and to ensure a precise identification of drivable area,(2)In order to obtain the spatial geometric information of the road under the intelligent driving scene,the binocular stereo vision is adopted to add spatial geometric information based on the RGB image,and establish RGB-D database to provide richer learning features for the recognition model.The semantic segmentation model is tested and verified,to ensure the model focusing on spatial geometry information.Experimental results show that for intelligent driving,the proposed shallow Ournet could recognize the drivable areas more precisely,quickly and robustly.The adding of spatial geometric features of roads based on RGB could improve the performance of semantic segmentation algorithm.