| At present,intelligent transportation video plays an important role in traffic safety and emergency response,but it is subject to light conditions,and the computer vision processing method relies heavily on video during the day.However,the actual environment is complex,and the illumination of the light at night is limited,which leads to the artificial surveillance mode.This paper,to deal with the problem,relying on "based on the deep learning charge lane recognition system of vehicle features" provincial projects,projects for night driving intelligent video detection study,it has important theoretical significance and practical application valueThe paper describes the difficulties in the intelligent detection of night driving.For example,it is hard to describe the car lamp and ground reflected light obstruct the detection.It also has light shading shadows,so that road signs are not easy to identify,etc.The common video processing algorithms are analyzed.such as,pyramids Lucas-Kanade and Blob algorithm.The prominent advantages of the feature fusion convolution neural network in target detection and tracking are discussed.Inspired by feature fusion of CNN algorithm,the system of the night light detection module,a nighttime vehicle detection module,and the parameters of driving demand analysis module,designed a series of traffic video detection algorithm is at night.It is sensitive to the change of night scene,and the headlights are not clear.The proposed bionic Faster RCNN nightlight detection method,and puts forward the night image enhancement method based on biology.It use the adaptive feedback of retinal horizontal cells and a bipolar cells at the center of the antagonism receptive field modeling to enhance the image contrast,brightness,prominent prospect light information.The enhanced image input to Faster-RCNN to training and learning.The experimental results show that the proposed light detection method has high robustnessFor the traditional vehicle detection method,the detection accuracy of video image is not high.Using the deep convolutional network layer to extract the rich features of the images,it use proposal network generate candidate boxes and filter out other targets,and fine-tune network to fine-tuning candidate boxes in order to generate precise location and recognition results.The paper use the framework of feature fusion method called EB,which realized the nighttime vehicle detection.localization and classification was optimized.The nighttime vehicle detection accuracy rate is improved.For night driving parameter analysis,high rate of vehicle tracking error and vulnerable to keep out problems are problems.we implement a variety of vehicle tracking algorithm for practical application analysis.Finally,we using the Siamese FC analysis speed,chi-square coefficient histogram matching analysis traffic flow.Implement the system’s driving parameters detection.So as to realize the detection of night driving parameters of the systemIn summary,the innovation points and features of the paper are·It proposes a Bionic image enhancement algorithm and train a Faster RCNN model to test nightlights This method improves the accuracy of the nightlight detection,up to 82.75%·It proposes a nighttime vehicle detection method,which is based on a deep learning framework called Evolving Box(EB).By optimizing the positioning of nighttime vehicle,the accuracy rate of nighttime vehicle detection was improved to 83.78%·It refers to camera calibration,the tracking detection algorithm,ROI area,implement driving parameters function at night such as precise counting and accurate speed... |
