Night Time Vehicle Detection for Adaptive Beam and Collision Avoidance Systems

This paper presents a novel and effective night time vehicle detection system for detecting vehicles in front of the camera-assisted host car. The proposed algorithm works for both oncoming vehicles (Head light detection) and preceding vehicles (Tail light detection). Image processing techniques are applied to the input frames captured by the forward looking camera fitted behind the windshield screen of the host car just near to the rear view mirror. The system uses a novel segmentation technique based on adaptive fuzzy logic, a novel statistical mean intensity measure and ‘confirmation - elimination’ based classification algorithm, and state of the art mutually independent feature based objects detection algorithm based on correlation matrix generation for the light objects identified in the scene. To distinguish true light objects from other false light objects present in the scene, it consists of shape and context based objects validation algorithm that uses properties like convex hull, collinear pattern and green - blue channels color variation. Detected objects are tracked based on matching of object properties. Distance and angle measurements are extracted for the objects using intrinsic camera parameter and geometry calculations. The proposed system is effective for multiple driver assistant applications like adaptive beam and forward collision warning. In multiple field tests, it is confirmed that system works efficiently in real time conditions. The system detects oncoming vehicle up to 1 km and preceding vehicles up to 250 m in ideal expressway.