Research On Traffic Flow Detection Technology Based On Machine Learning

With the rapid economic growth,the domestic automobile holdings have continued to grow rapidly.The large number of cars has brought tremendous pressure on urban traffic,and there have been a series of problems such as congestion.Blockages and frequent traffic accidents have become A chronic disease that restricts social progress and economic development in various regions.Problems such as improving road capacity,reducing traffic congestion and reducing traffic accidents need to be resolved.The emergence of urban smart transportation can alleviate the pressure of urban traffic to a certain extent.Traffic flow detection based on surveillance video is an important part of intelligent transportation.This paper focuses on the research of multi-target moving vehicle detection and tracking technology in the field of traffic flow detection.Video-based moving target detection and target tracking have strong practical value,mainly used in video surveillance,video image compression,intelligent transportation,robot navigation,medical image analysis,industrial detection and other fields.Due to the diverse environments of surveillance video,different scenarios,diverse targets,and many interferences,it is necessary to study the detection and tracking technology of multi-targets in motion to better apply traffic monitoring video to detect traffic flow.In this paper,based on the basic theory of multi-target detection and tracking,a video-based motion detection and tracking method is proposed.Based on the complex environment of urban traffic conditions,the depth learning-based Mask R-CNN(mask area CNN)algorithm identifies the vehicle contour.The Kalman filter is then used to perform motion tracking and statistics on the vehicle targets in the video sequence.Experiments are carried out using video samples collected by text.The experimental results show that the proposed method can effectively detect scenes during daytime,nighttime and congested road conditions,and can accurately detect and track vehicle positions during daytime,nighttime,and crowded roads.The average detection accuracy in the scene test reaches 95.45%,which has good accuracy and robustness,and basically meets the requirements of actual traffic flow video detection.The main work and innovations of the thesis are as follows: 1)Based on the existing public vehicle dataset,collect and add up to 10,000 vehicle images,especially domestic models.The improved vehicle dataset covers network vehicle images and vehicle images from actual surveillance video extraction of urban traffic,and collects traffic videos of relevant test sections of different scenes as experimental data.The test image dataset includes day,night,night.Different road conditions such as rainy days and traffic congestion;2)Designed a multi-target vehicle detection model based on Mask R-CNN.Through the segmentation of the vehicle area and the background area,the background interference is removed,and the foreground image is fine-grained.After corresponding improvement and continuous debugging,the model is optimized and the validity of the model is improved.After training,the mAP of the target detection reaches 59.37,and the accuracy of the position detection of the natural image in the data set reaches 100%,and the area of interest in the actual monitoring video is within the area.The accuracy of vehicle position detection reached 98.3%;for multi-label classification tasks,the average correct rate of natural scene images was 93.70%;the average correct rate of daytime scene images intercepted by surveillance video was 94.91%,and the average correct rate of night scene images was 88.18%.3)Traffic flow detection algorithm based on Mask R-CNN vehicle detection and Kalman moving target tracking.For urban traffic monitoring video,the accurate vehicle profile detected by Mask R-CNN is used to perform multi-target real-time tracking with Kalman filter,which improves the counting accuracy and real-time performance of multi-target vehicles under complex traffic conditions.