This paper details about a framework for preceding obstacle detection for autonomous vehicles or driver assistance systems (DAS). Detection and tracking surrounding moving obstacles such as vehicles and pedestrians are crucial for the safety of people and autonomous vehicles. The proposed system uses four main stages: input, feature extraction, obstacle detection, and output of system. The system has input in video format which will be converted into frames. Feature extraction stage extracts LBP+LTP feature and Gabor feature separately. With the help of training set and extracted features, CNN will detect the obstacle in the frame. The system with LBP+LTP feature in cooperation with CNN gives 86.73% detection rate and 4% false alarm rate and the system with Gabor feature in cooperation with CNN gives 86.21% detection rate and 0.27% false alarm rate. From the proposed system, we can conclude that computer vision in combination with deep learning has the potential to bring about a relatively inexpensive, robust solution to autonomous driving. © 2019, Springer Nature Singapore Pte Ltd.