Vehicle to vehicle distance measurement is one of the main tasks of the Advanced Driver Assistance System (ADAS). To measure the distance between the subject vehicle and the target vehicle from images, this study proposes a machine learning approach aiming at developing algorithms that can learn and create statistical models for data analysis and distance prediction. The proposed model consists of the YOLO (you look only once) method for vehicle detection task and the supervised neural network learning algorithm for distance estimation. Therefore, this research is divided two stages. In the first stage, the purpose is to produce bounding boxes of multiple objects within the image and generate ground truth of distance between vehicles from KITTI dataset. The KITTI dataset is a popular dataset which can be used for vision algorithm testing of ADAS. It contains a large number of stereo image pairs captured from a car driving in an urban scenario and also provides sparse depth data matched with the stereo vision. On the basis of the first stage, information pertaining of the vehicle objects and their bounding boxes are used as input features for neural network model to learn the implicit distance relation with preceding vehicles in the second stage. The experimental images from real road scenarios are extracted the public KITTI dataset. The results of the quantitative and qualitative comparisons on the KITTI dataset show that the proposed neural network model can effectively predict vehicle distance.

Login to Download PDF

Please login with your Paper ID and password to access the full PDF.

🔑 Login to Download