Vehicle Classification in Nighttime Using Headlights Trajectories Matching

Abstract

Vehicle detection and classification is an essential application in traffic surveillance system (TSS). Recent studies have solely focused on vehicle detection in the daytime scenes. However, recognizing moving vehicle at nighttime is more challenging because of either poor (lack of street lights) or bright illuminations (vehicle headlight reflection on the road). These problems hinder the ability to identify vehicle’s shapes, sizes, or textures which are mainly used in daytime surveillance. Hence, vehicles’ headlights are the only visible features. However, the tracking and pairing of vehicle’s headlights have its own challenge because of chaotic traffic of motorbikes. Adding to this is various types of vehicles travel on the same road which falsifies the pairing results. So, this research proposes an algorithm for vehicle detection and classification at nighttime surveillance scenes which consists of headlight segmentation, headlight detection, headlight tracking and pairing, and vehicle classification (two-wheeled and four-wheeled vehicles). The novelty of our work is that headlights are validated and paired using trajectory tracing technique. The evaluation results are promising for a detection rate of 81.19% in nighttime scenes.