Research On Front Vehicle Detection And Tracking Based On Multiple Features

The road traffic problems have become serious in our nation by threatening people's life and assetting with the growing number of automobiles. Driving-aid system, which is aiming at improving the traffic situation through high technology, has gotten more and more attention as a significant part of the intelligent transportation system. The detection of the vehicles plays an indispensable role in the vehicle-detection. The real-time driving environment is obtained through a vehicular camera which is used to detect and follow the vehicles in front of the host one.The algorithm presented in this brief is improved based on various papers concerned home and abroad. The paper is organized as follows:1. This paper presents a new real-time and robust approach based on the combination of the vehicle shadow and characteristics of the road gray level and gratitude to detect forward vehicle. Define a effective area in the image and calculate the gray values in segments, based on which detect the vehicular shadow through the difference of gray values between the shadow and the road. Build a region of interest of the vehicle.2. The method presented in this paper is based on the fusion of the texture, shape and edge symmetry. The texture extraction is based on improved differential box counting acquiring a better result. The shape feature extraction is based on the matching degree of edge images. The symmetry feature is extracted by counting the symmetry measure of the vertical edge projection. After counting the three kinds of characteristic values,we found the Multi-feature fusion decision function by caculating each feature value with variance measure. Then we use the decision function determine whether the ROI contains vehicle.3. Vehicle tracking based on Kalman filtering. Kalman filtering is used for vehicle following based on vehicular position and size in the image. Update the observing vector in the possible area by the features of the vehicular edges and shadow; testify the following objects through the NMI feature and the entropy of the vehicle area.4. Program a vehicle detecting and following system on the platform of Visual C++6.0 in windows. Experimental results using video shows that the method presented in this paper can detect vehicles in real time and high accuracy.