Nighttime vision-based car detection and tracking for smart road lighting system

The objective of this paper is to detect cars in nighttime videos for controlling the illumination of level of road lights, thereby saving power consumption. We present an effective method to detect and track cars based on the presence of head lights or rear lights. We detect the headlights and rear lights of the cars using their luminance value, geometry and location on the road. The algorithm is validated using multiple test sequences. The test sequences are manually annotated to identify the position of the car lights in the video frames. The method is robust against poor and variable light condition of the scenes. The results show that in sequences where head lights occur without coding artifacts the success rate is above 90%. However, the precision of the algorithm degrades if the car is moving in a irregular trajectory. This happens, for example if there are reflections that follow a car for a long duration.