Nighttime Vehicle Detection for Heavy Trucks

This paper presents a method for detecting vehicles at nighttime, particularly for an application in heavy trucks. Researchers suggested detecting vehicles at nighttime based on symmetry of taillights, or by training a classifier. The headlights of heavy trucks are very bright which causes that taillights of a vehicle in front appear as being asymmetrical. The bright headlight also defines disturbing information that affects the training of a classifier. We propose an improved threshold algorithm that can effectively remove most non-taillight regions and strengthen the shapes of taillight pairs. Positive and negative samples are extracted from these thresholded pre-processed images to train a vehicle classifier by using Haar-like features and AdaBoost. We then detect vehicles in these pre-processed images. Experiments are performed using two alternative methods. Results show that our method, i.e. combining threshold pre-processing and training a classifier, is more accurate and robust than the other two methods.