Night Time Detection of Vehicles

Abstract

This paper presents results of a study carried out to count and classify vehicles in video sequences of traffic scenes captured from a fixed digital camera under night conditions. The vehicle identification has been carried out using the headlights, which may be the only key feature available to identify vehicles under low light conditions through visible light. Due to lack of features, the vehicle classification was limited to two classes as heavy and light. The development process incorporated a number of pre-processing steps; background subtraction, which was used to extract moving headlights from the static background, followed by grayscalling, thresholding and noise filtering, which were carried out to help with the accurate identification of headlights. The system was build with the assumption that all vehicles switch on their headlights at night-time and all headlights are working properly. The current accuracy of the system for counting vehicles is 89% and that of vehicle classification is 88% for heavy vehicles and 90% for light vehicles. Present accuracy has the potential to improve with further studies.