Abstract
In today’s world, many car accidents occur due to driver’s behavioral errors. Some of these accidents can be avoided by using intelligent driver assistance systems. Such systems must include a collision detection component. Different sensors can be used including vision sensors to detect whether there is a collision risk with other objects or vehicles. The main aim of this project is to design a system which uses a camera mounted on the test vehicle that will detect any preceding vehicle in the image frame. Depending upon the distance travelled by both vehicles in a particular time interval, this system can predict whether the vehicle in the frame will collide with the test vehicle. Ego motion and optical flow are some of the techniques used to detect vehicles. Ego motion is used to determine the 3D position of the camera mounted on an object relative to its surrounding whereas optical flow is used to detect moving vehicles by capturing image frames using optical flow vectors and comparing them with the model vectors depicting expected vehicle motion. In order to detect objects in the image frames, it is necessary to map the image frame coordinates to the world coordinates. With the help of this mapping, the coordinates of the object in the real world can be found out. Also, as the image is made of pixels, it is necessary to find out the relation between the image coordinates and the pixel coordinates which is done by camera calibration. Finally, by converting the video captured by the camera into image frames, a collision detection method is used to predict if the cars are going to collide by calculating the distance travelled by the cars in the particular time intervals. In conclusion, this project aims to predict collision of vehicles in image plane by analyzing the images obtained from the video captured by the camera.