Obstacles Recognition In Front Of The Vehicle Based On Convolutional Neural Network

Road traffic safety issues become increasingly prominent with the increasing of vehicle population. In order to reduce traffic accidents effectively, driver assistance system (DAS) has been paid widespread attention. And obstacles recognition in front of the vehicle is one of hotspots and key technologies in DAS. It can proactively identify and effectively reduce risks on the road by various sensors, thus it has been widespreadly concerned by governments, enterprises and academes all over the world.As RBFNN has the superiority of simple structure as well as rapid convergence and strong expression ability, while convolutional neural network (CNN) is partially connected and with sharing weights and down sampling, we proposed a method of obstacles recognition in front of the vehicle basing on RBF’s convolutional neural network (RBFCNN). And we did the experiment using the samples established and proved the efficienty of the network structure established in this study.For the reason that online learning is a real-time learning with easy execution and a small quantity of memory space, here we designed an online RBFCNN. Then trained and examed the network by handwritten digital database MNIST. In the mean time, a semi-automatic sample collection methord is designed imitating the real highway, which laid the foundation for the compeletely online training.Then, we got a training set compared to the high leveled and structured road to be used in the experiment of obstacles recognition in front of the vehicle. And we trained the improved convolutional neural network with the training set and got the final network structure. At last we did the final experiment of obstacles recognition and improved our method is effective and highly generalizationed and extended.