Junction Detection Module In Intelligent Transportation System For Wuhan

With the progress of the times and the development of science and technology,China's urban monitoring system has seen a diversification of functions and artificial intelligence from the original human inspection to the later integrated monitoring to the current smart city and intelligent transportation.Artificial intelligence is a very important direction in the industry.Since deep learning and convolutional neural network emerged in 2012,it has brought new changes to the entire industry,injected new vitality,and created the entire industry landscape.With profound changes,traditional veteran monitoring and outsourcing companies tend to explore and develop traditional methods as prior knowledge of existing deep learning models.Emerging companies such as Qu Shi and Shang Tang tend to develop new deep learning models to achieve Public security department design requirements.Since last year's Skynet system launched a new face recognition system,the problems that traditional methods cannot solve through deep learning and artificial intelligence have become one of the mainstream development directions of intelligent traffic and smart public security systems.This article mainly uses deep learning and artificial intelligence technology to research and explore the needs of Wuhan's traffic management department.The main research direction is to identify pedestrians and vehicles and to identify vehicles under the complex environment and different interference conditions.Will courtesy pedestrians make judgments and submit the results of the judgement and the three pictures of the evidence chain to the existing traffic control system.At the same time,judge the remnants of the intersection and submit the two pictures of the evidence chain to the system.In the process of researching the intelligent transportation system,based on the training data provided by the company,the training library of the deep learning model is divided into two parts to train,one is the pedestrian detection training,the other is the vehicle detection training,according to the public security The requirements of the relevant standards provided by the ministry divided the training of the vehicle into three parts: the front,the body,and the parking space.The training model uses the YOLO model.After about 30,000 batches,training results are obtained,and the accuracy rate is over 85%.After the model is identified,the positions of other boxes will be returned to the video image.Because the positions of cameras and crosswalks at each intersection are relatively fixed,in order to save the calculation power and speed up the operation,we manually calibrate the crosswalk area as the entire area.The ROI identified by the system is identified and identified in the ROI area.First,the pedestrian's coordinates are taken as the reference point,and the pedestrian's movement direction is recorded as the marked position.The area in front of the pedestrian is set as a no-line zone.Vehicles are not allowed to enter.The area behind the pedestrian is a feasible area,and no detection is performed.At first,the pedestrian's position was determined,then the feasible area and the no-line area were designated,the vehicles entering the forbidden area were tested,and the illegal pictures were sent to the police system.Relic detection is also used to monitor the ROI area in the video,using the frame difference method and the hybrid Gaussian model to calibrate the remnants and transfer the remnant images to the police system.