Danger Detection from Head Light Glare through Vision Estimation

摘要

Driver's night drive vision can be seriously reduced by opposite headlight glare especially in mountainous roads. Though there are methods for reducing opposite headlight glare effects, these techniques usually deteriorate night vision in normal state. Therefore, these techniques must be deployed if and only if a certain danger from the opposite headlight glare become really danger. The method for detecting this danger is proposed based on ordinary image or video captured device as anyone who own car can buy, using properties in every still color image to extract light source and eliminate unconcerned objects such as some reflects on pavement, background color from own light beams, and then find the most area in picture to determine whether it is dangerous glare or not. The result measurement performs by manually to find trend of percentage area for warning. The proposed method is evaluated based on recorded videos from various scenarios. The test implies good result based on no probability of negative fault error with only low of positive fault error probability.