Energy constrained forward collision warning system with a single camera

This paper has conducted the research for developing Forward Collision Warning System (FCWS) with a single camera using energy constrained methods rather than using high computational methods such as optical flow and machine learning. To estimate collision time between ego-vehicle and target object, we set them with 2D modeling; using time, the relative distance and the relative velocity. The proposed algorithm has three stages; generating region of interest, locating target object and estimating time to collision. In the first stage, we used lane detection with Random Sample Consensus (RANSAC) to generate region of interest. The second stage locates target object with shadow detection and edge information which have lite computational cost. Lastly, in the third stage, time to collision is estimated from the differences of the calculated relative distance for each frame. To show the performance, the algorithm was evaluated by comparing the result of the relative distance from the proposed algorithm with the relative distance from ground truth.