Collision Detection through Optical Flow

Abstract

In this paper, an analysis of collision detection via optical flow is presented. The main objective of the work is to detect and initiate a warning signal when there is a non-moving object or a moving object (such as a pedestrian) obstructing the path in which a vehicle is moving. The warning signal is given only if there is a risk of a collision. In order to gather the input video streams to the simulation, a camera with a resolution of 1280×720 pixels was used. Since the processing time increases with resolution, the actual processing was carried out at a lower resolution of 160×120 pixels. The colour segmentation was carried out through RGB colour space. Simulink was used as a platform for the development and testing the simulation. Using the Simulink blocks, video pre-processing, image segmentation, optical flow calculation and thresholding were carried out. The stability of the video is one of the main concerns in this research. For videos taken when the vehicle is moving at high speed, the stability became an issue. The developed system tested successfully at speeds below 10 kmh-1 to detect stationary objects and pedestrians crossing the roads.