Combining CNN And MRF Road Detection Methods

With the improvement of economic development level,the number of cars continues to rise and the incidence of traffic accidents is also increasing.The intelligent driver assistance system can reduce the incidence of traffic accidents by reminding or guiding drivers.Road detection based on vision is the key to the intelligent assistant driving system,which can not only provides clues for obstacle detection,but also help vehicle path planning.Because the road scene is complex and changeable,the road detection result is easily influenced by lights,water spots,shadows,obstacles and other factors.Therefore,the road detection algorithm for complex scene has become the research hotspot.To improve the robustness and accuracy of road detection methods in complex environments,a new road detection method based on CNN(convolutional neural network)and MRF(markov random field)is proposed.First of all,the original road image is segmented into super-pixels of uniform size using SLIC(simple linear iterative clustering)algorithm,which can not only reduce the computational complexity of the subsequent processing,but also help to extract local features and keep target boundary information.Then,aiming at the problem that the traditional classifiers need to design features artificially and its poor adaptability in complex scenes,we utilize the CNN to learn the essential features of dataset from a large number of samples automatically.An appropriate network structure is designed and the network model is trained on the sample set based on super-pixel segmentation.The trained network model is applied to classify the road region and the non-road region.Finally,to improve the accuracy of the detection results,according to the relationship between the super-pixel and its neighborhood,the MRF model is used to optimize the classification results of CNN.Quantitative and qualitative analysis are performed on the public dataset.The experimental results show that the proposed method can effectively detect the road region ahead of the vehicle and has good accuracy and robustness in complex scenes.