Research On Vehicle And Lane Detection Algorithm In Night Vision Environment

Vehicle detection and lane detection play a crucial role in the design of automated driving systems(ADS),and significant progress has been made in recent years.At present,vehicle detection and lane detection algorithms have achieved good detection results in the daytime environment,but there are still many research difficulties in the detection problem in the nighttime environment.Due to the light interference on the road at night,the lights in some road areas are too dark,and the lights of passing vehicles are too bright,the vehicle characteristics are not obvious,and the vehicles at night block each other,so it is difficult to detect small target vehicles in the distance.Vehicle detection still presents considerable challenges.In addition,in the nighttime lane detection scene,the lane will be affected by conditions such as incomplete,damaged,strong light and dark light,resulting in poor detection results.In order to solve the above-mentioned problems in the process of vehicle detection and lane detection at night,this paper starts from two parts: nighttime vehicle detection and nighttime lane detection,and proposes a vehicle detection algorithm and lane detection algorithm with higher accuracy.The main contents of this paper are as follows:(1)Aiming at the problems existing in vehicle detection at night,a vehicle detection algorithm in night vision environment is designed.First,the Attentive GAN network is used to process the night road images,which consists of a generator network with an attentive module and a dual discriminator network—a global discriminator and a local discriminator network to balance the local over-dark and over-exposed areas in the night road image.The processed images are further detected by the improved Faster RCNN network,that is,on the basis of the original Faster RCNN,the classification and regression parts are improved.In order to obtain higher detection accuracy,multiple local regression is used in the regression branch to predict multiple bounding box offsets,and finally a more accurate prediction box is obtained through calculation.The improved classification branch utilizes discriminative features to obtain higher classification scores.The experimental results show that the vehicle detection method in this paper shows good detection performance in a series of complex scenes such as extremely dark environment,glare,and occlusion.(2)Aiming at the problems of low detection accuracy and poor lane fitting effect in lane detection at night,a lane detection algorithm in night vision environment is proposed.In the lane detection network,the Attentive GAN network is used to process the night road image,highlighting the lane features in the night road,and Res Net-18 network is used to extract the lane features,and SCNN algorithm is used to detect the lane.Experiments show that the algorithm proposed in this paper has good performance of lane detection at night.