Research On All-weather Light Recognition Technology For Assisted Driving

The assisted driving system can perform assisted driving function by sensing information in the surrounding environment and monitoring the driving state of the driver when the vehicle is moving.The lamp language is important for the vehicle to make driving decisions during the driving process.Accurately identifying the lamp language information of the preceding vehicle can help the assisted driving system to fully utilize its auxiliary functions,which is of great significance for improving driving safety and vehicle intelligence.Lamp language recognition technology can be divided into two situations: daytime and nighttime.This paper conducts in-depth research on the all-weather lamp language recognition problem.The main work contents are as follows:1.According to the vehicle detection problem under daytime driving conditions,two kinds of vehicle detection methods based on machine learning were studied and compared,and the convolutional neural network was selected as the vehicle detection method.The Haar feature combined with the Adaboost cascade classifier detection method and the convolutional neural network vehicle detection method were used to detect the vehicle in the video.Experiments show that the detection method of the convolutional neural network is more accurate,but its efficiency is lower.Therefore,the convolutional neural network is decided to use to implement vehicle detection.2.For the nighttime driving,the glare problem existing in the image collected by the driving recorder was studied,and a glare removal method was proposed,which could effectively remove the glare in the image.The dark channel prior de-fog method was used to remove white glare,and an accelerated calculation method was proposed to ensure the real-time performance of the algorithm.And the method of removing glare of a specific color was proposed,which could effectively remove the specific color glare.3.For vehicle detection in night driving situations,a pair of position lights on the same car were used to locate the vehicle.First,use the positional relationship of the position light to establish and constrain its position indicator,and then use a tracking algorithm to track the detected pair of position lights.4.For the two situations of daytime and nighttime,the corresponding lamp language recognition method was proposed.For daytime driving conditions,the vehicle was first detected,then the area of interest of the headlights was established,and the taillight position was detected by color division method,and its taillight status was determined by combining the brightness characteristics.Finally,the lamp language recognition rule was formulated to realize the daytime driving situation.5.For night driving conditions,the brake lights were detected by matching the high position lights.The light splitting method was combined with the historical information to judge the turn signal,then,the light word recognition in the case of night driving was realized.