PROJECT TITLE :

On-Road Collision Warning Based on Multiple FOE Segmentation Using a Dashboard Camera

ABSTRACT:

Various accidents can be avoided if drivers are alerted simply a few seconds before a collision. But, collision prediction is difficult because of high computational loads, complicated background muddle, and nonstationary sensors. Active sensors, like ultrasonic, radar, and laser, are expensive and will cause interference problems in significant traffic. Thus, this paper explores the chance of a visual collision-warning system solely employing a single dashboard camera that is currently widely obtainable and straightforward to put in. Existing vision-based collision-warning systems specialise in detecting specific targets, such as pedestrians, vehicles, and bicycles, based on statistical models trained earlier. Rather than hoping on these prior models, the proposed system aims at detecting the overall motion patterns of any approaching object. Considering the fact that each one motion vectors of projecting points on an approaching object diverge from a purpose known as focus of growth (FOE), we construct a cascade-like call tree to filter out false detections within the earliest attainable stage and develop a multiple FOE segmentation algorithm to classify optical flows to distinct originating objects based mostly on their individual FOEs. Any analysis is performed on objects in a high-risk space referred to as the danger zone. Tracking steadiness is examined, and therefore the time-to-collision (TTC) is estimated to evaluate collision risks.