Vision-Based Integrated Techniques for Collision Avoidance Systems

Vehicle detection and lane detection form integral part of most advanced driver assistance systems (ADAS). While most systems employ individual algorithms for the two tasks, synergistic use of the two algorithms is less explored. In this chapter, we introduce an integrated approach called efficient lane and vehicle detection with integrated synergies (ELVIS) that exploits the inherent synergies between lane and on‐road vehicle detection to improve the overall computational efficiency without compromising on the robustness of both the tasks. While the lane information is used to localize the possible candidates for vehicle detection, the detection output from the vehicle detector is used for customizing the parameters for lane detection. Such a synergistic approach not only increases the accuracy of both the tasks but also enables reduction in computational cost in the context of embedded computing architectures. Detailed evaluations show that the vehicle detection component of ELVIS shows at least 50% lesser false alarms with equal or better detection rates, and reducing the computational costs by over 90% as compared to state‐of‐the‐art vehicle detection methods. Similarly, the lane detection component shows more reliable lane feature extraction with average computation costs that are at least 35% lesser than existing techniques.