Vehicle feature recognition based on the frontal lamp pattern descriptor

In this paper, a light pattern descriptor(LPD)is proposed to describe the vehicle features for vehicle type recognition. Based on the LPD, the frontal light position distribution of vehicle is utilized as features of the vehicle, and it is composed of horizontal and vertical relative distance of frontal lamps, which are easy to be computed. This descriptor is robust because it's invariant to rotation and scale, and is insensitive to noise. This feature makes LPD very useful in traffic recognition because the resolution and direction of view of images are variant in reality. In many cases, for different images, the scales of the vehicle may be also different. So the stability of features is very important. Experiments verify that this descriptor is effective and accurate in vehicle feature extraction and type recognition.