| The ADB adaptive driving beam(ADB)system based on LED matrix headlights has become a technical trend for nighttime automotive safety lighting.the ADB system acquires the road images ahead through cameras,detects the vehicles in the images using nighttime vehicle detection methods and outputs the location information,and turns off the LED lights in the corresponding area based on the location information.However,the traditional nighttime vehicle detection method is based on extracting a single feature of the vehicle lights,and the lights are easily confused with the ambient lights,thus causing misjudgment,and the detection accuracy and real-time performance are not high.In this paper,we propose a nighttime vehicle detection algorithm based on optimized Light Net+Res Net50+CBAM+Faster RCNN,in which optimized Light Net is used to enhance the brightness of the vehicle area in the image,Res Net50+CBAM is used to extract the vehicle outline as well as the headlight features,and Faster RCNN is used to identify the vehicle features and The Faster RCNN is used to recognize vehicle features and output location information.The research design of the control module and lighting module of the ADB system is completed based on the detection model,and the hardware experimental platform is built to test the detection model and the performance of the ADB system.The main research contents are as follows:(1)Theoretical analysis of the components of the general ADB system and the function,opening or activation conditions and working principle,proposing an ADB system architecture based on visual night vehicle detection;In order to achieve the function of the Night Vehicle detection of the ADB system,the design architecture of night vehicle detection module is established;To achieve an adaptive adjustment of the ADB system,establish a control and lighting module design architecture.(2)Based on the design architecture of the night time vehicle detection module,the related theory of night vehicle detection is carried out.In response to the image illumination in the night vehicle detection,it is difficult to obtain the vehicle characteristics,and the optimization Light Net image enhancement network is proposed.The L2 norm is added to the algorithm optimization method,and the use of adaptive GAMMA correction,enhance the vehicle area brightness;In response to the low problems of the night-time vehicle detection accuracy,the Res Net50 + CBAM’s backbone feature extraction network is proposed,and the residual network Res Net 50 is used to extract the lamp and vehicle contour characteristics,and the CBAM module can enhance the target channel and space in the residual structure diagram.Feature information,improve detection accuracy;The night-scale vehicle detection algorithm based on optimizing Light Net + Res Net50 + CBAM + Faster RCNN is proposed to realize the detection of night vehicles;In response to the real-time problem of night-time vehicle detection model,reduce model parameters and volume using model lightweight method,improve detection real-time;Building a Night Vehicle Detection training data set through a data enhancement method,training the algorithm model to obtain the final detection model.(3)The control module and the lighting module are studied based on the control and lighting module design architecture.In order to realize the corresponding area light according to the vehicle position,this paper proposes an LED optical partition scheme to divide the lighting area into 8 pieces,each piece is responsible for lighting by a single LED light;The core heat dissipation and lens design of the LED lamp is theoretically,the LED lens of the air-free heat and glass is suitable for matrix LED headlights,and 8*8-LED light board and 8 sets of COBS LEDs are proposed as ADB.System LED matrix headlights;Integrated LED optical partitioning scheme and vehicle location information,building partition control schemes,with two LED matrix headlights of the light module,to construct a partition control program design and partition control circuit design based on Arduino microcontroller;A light correction method based on Deep Sort target tracking is proposed for the light deviation.(4)Construction of the hardware experiment platform is completed based on the ADB system design architecture.Performing a night vehicle detection model and ADB system performance test experiment.In the Night Vehicle Detection test set,accuracy and detection real-time experiments are performed.The MAP value of the night vehicle detection algorithm is 91.62%,and the number of fps value of the frame per second is 31,and there is almost no occurrence of misjudgment,and in various types The vehicle can accurately detect the vehicle under the scene.Testing the ADB system,under all kinds of driving scenarios,can meet the light of the LED matrix large lamp corresponding to the LED matrix large lamp according to the detected vehicle position to meet the requirements of the ADB system.The ADB system proposed in this paper provides theoretical guidance and practice application foundation for subsequent ADB system research and production. |
