A Low-Cost and Fast Vehicle Detection Algorithm With a Monocular Camera for Adaptive Driving Beam Systems

The adaptive driving beam headlamp system is helpful to solve the traffic safety problems caused by the abuse of high beams at night. The fundamental problem for the adaptive driving beam system is to detect and track the vehicles at specific night environment for light shape control. This paper proposes a new and efficient algorithm for detecting and tracking based on video collected by a low-cost camera. The related automobile regulations are analyzed for classifying the application scenarios of the adaptive driving beam system and summarizing the performance requirements of the algorithm. A standard CMOS camera mounted behind windshield captures the test video from different scenarios. Some preprocessing approaches are designed to optimize the captured video so that the algorithm can work independently on specific camera. The color and morphological characteristics of the rear lights are utilized to extract the rear lamps of the vehicle. The symmetry of rear lights is checked by a correlation coefficient method to pair the rear lamps and determine the vehicle ahead preliminary. Then, the Hungarian algorithm and Kalman filter are performed to track the multiple occurrences in two consecutive frames and correct the detection results. Finally, an estimation method is given for calculating the vehicle position in real world. The experiments are designed according to referred regulations and the test video is obtained from a low-cost camera mounted on test vehicles driving in specific scenarios. The experimental results demonstrate that the algorithm can get high detection rates and adaptability to the working condition of adaptive driving beam system. Moreover, the proposed algorithm has low time cost and can be applied in embedded devices of the vehicle.