Reserch On Low Illumination Vechile Detection Method Based On Deep Learning

Autonomous driving technology can improve driving safety,reduce casualties and property losses.In the process of autonomous driving,it is first necessary to perceive the driving environment around the vehicle.Among them,collecting images through the onboard camera is an important perception method,but in low illumination The environment is likely to cause the quality of the collected images to decrease,which in turn will affect the safety of autonomous driving.To this end,this paper has carried out research on lowlight vehicle detection methods based on deep learning,and selected corresponding models in vehicle detection related theories through investigation of deep learning,and designed S-Enlighten GAN low-light image enhancement model based on deep learning.The vehicle target detection model based on deep learning realizes the vehicle detection,and the experiment and analysis of the vehicle target detection model are carried out.The specific research content of this paper is as follows:(1)The vehicle detection theory based on convolutional neural network,the vehicle target detection theory based on deep learning,and the image enhancement theory of generative confrontation network are studied.The principles of vehicle detection tasks based on convolutional neural networks are studied,and then the characteristics of singlestage and two-stage target detection modes are studied,and finally the characteristics of generative adversarial networks are studied.(2)Designed S-Enlighten GAN low-light image enhancement model.To improve the quality of images collected in low-light environments,an adaptive filter network module is designed to solve the problem of checkerboard effect in image enhancement methods based on deep learning,which effectively suppresses the generation of checkerboard effects;for current image enhancement the paired data set used is difficult to obtain,and the artificially synthesized image is not real enough.Construct a non-paired low-light image enhancement data set,let the S-Enlighten GAN model learn the characteristics of the light distribution in the image,and finally effectively realize the low-light image Enhancement.Finally,the low-light image enhancement method proposed in this paper is compared with several other commonly used methods,which proves the superiority of the S-Enlighten GAN low-light image enhancement algorithm.(3)The design of a vehicle target detection model based on deep learning realizes the rapid detection of vehicles.To detect vehicles quickly and accurately during driving,a vehicle target detection model based on deep learning is first designed to achieve vehicle detection.In view of the high requirements of vehicle detection in real-time,a structured and unstructured phase is designed.The combined model pruning strategy greatly improves the detection speed of the model.(4)Launch vehicle target detection model experiment and analysis.The vehicle object detection method proposed in this paper is compared with several commonly used methods from the three aspects of detection accuracy,detection speed and visualization results,which proves that the vehicle object detection method proposed in this paper has more advantages.The experimental results show the effectiveness of the S-Enlighten GAN low-light image enhancement algorithm and vehicle target detection method proposed in this paper.It can improve the quality of images collected by self-driving cars at night,improved perception and driving safety.