Nighttime Motorcycle Detection For Sparse Traffic Images

Abstract

Traffic accidents mostly occur at night. It is understandable since at night, we have low visibility. Effort to reduce accidents at night have been reported by developing tools to detect nearby vehicles to avoid crashes. However, most of them worked only on detecting cars. They did not focus on motorcycle due to its lack of a pair of lamps lighting features usually utilized in detecting cars. Meanwhile, motorcycle consists of a rider, a taillight, and area surrounding license plate in the nighttime traffic images. To get these properties, we propose four features (red pixel, edge pixel, edge ROI, and active contour) that are extracted from red and edge maps. The red map is used for recognizing the spreading out of taillight on the image. The edge map is used to recognize the rider, back part of the motorcycle, and the whole curve of object. To see the effectiveness of our features, we selected 3 commonly used classifiers (Artificial Neural Network, Decision Tree Algorithm, and Support Vector Machine) in the experiment. The result shows that some classifiers have achieved more than 80% accuracy rate.