Research On Key Technologies For Expressway Vehicle Detection And Trajectory Prediction With Optical Distributed Acoustic Sensing

Compared with the traditional vehicle detection technologies,the traffic monitoring technology based on Optic fiber distributed Acoustic Sensing(DAS)has the advantages of continuous distributed information sensing,strong anti-electromagnetic interference ability,low cost so as to meet the needs of the whole time-space traffic information collection of smart expressway.However,the current technologies can not adapt to the expressway environment with strong noise interference and high real-time requirement.The main technical bottlenecks are the lack of anti-interference and low delay data processing algorithms,as well as the lack of actual expressway DAS data samples.Considering the practical problem of expressway traffic detection,this paper studies the system and algorithm of the vehicle detection and trajectory prediction on expressway by taking advantages of distributed optical fiber sensing technology.The research results and innovations of this paper are summarized as follows:1.This paper presents a new expressway vehicle detection scheme based on distributed optical fiber sensing technology.The free fiber cores in the highway communication fiber and the DAS detection equipment based on(37)-OTDR are used to collect the ground vibration signal iccuring from the driving vehicles.The vehicle detection and trajectory extraction are carried out by DAS signal denoising and vehicle trajectory enhancement so as to realize long-distance vehicle tracking.In order to verify the effectiveness of the method,a large number of vehicle DAS vibration signals are collected in the real highway environment of two-way three lanes with a length of 6000 meters to form a data set HW6.The proposed algorithms are tested and verified on this data set,and compared with the experimental result of other algorithms.2.DAS system is affected by the noise in the complex environment of the actual expressway,so the signal-to-noise ratio of the collected signals is very low.This paper studies the time-frequency domain characteristics of DAS vehicle vibration signal,and proposes a vehicle trajectory enhancement and extraction algorithm based on DAS.The vehicle trajectory enhancement algorithm based on S transform can enhance the ground vibration signal generated by vehicle driving,but suppress the vibration signal and noise generated by the operation of vehicle engine and other components.On the basis of vehicle driving trajectory enhancement,a Radon transform method of multi-vehicle vibration signal separation is presented,which can extract vehicle driving trajectory without supervision.The experimental results show that compared with the method based on wavelet and Noise2 Noise,the vehicle trajectory is more obvious after being processed by this S-transform based algorithm,and compared with the method based on Hough transform,more local peaks is accumulated and clustered in Radon domain,so it is easier to separate multi-vehicle trajectories.3.In the traffic detection method based on DAS,the supervised machine learning algorithm can not obtain a reference signal with high signal-to-noise ratio,while the denoising performance based on classical time-frequency domain transform is limited by the number of time sampling points and time sampling frequency.In order to solve these problems,this paper presents a vehicle trajectory enhancement algorithm based on x-transform and convolution neural network Noise2X-Net(X represents any classical denoising algorithm with long time window).Noise2X-Net is a training pair composed of noisy signals and enhanced signals based on X-transform to learn the mapping from noisy signals to enhanced signals,so as to reproduce the vehicle trajectory enhancement performance achieved by X-transform under the condition of short sampling time and low sampling frequency.Experiments show that by using signals collected within extremely short time windows of 100 ms,an insufficient time for the processing of existing denoising algorithms,the algorithm yields a satisfactory denoising performance.4.Vehicle trajectory prediction is the basis of expressway traffic situation prediction and intelligent driving surrounding environment prediction.In order to fill the gap of the research of DAS vehicle trajectory prediction,this paper constructs a DAS data prediction model Past2F-Net by using LSTM algorithm,which can input the DAS vehicle vibration denoising signal into the model to obtain the predicted DAS signal.After extracting the vehicle trajectory by exploiting the algorithm based on Radon transform,the vehicle trajectory prediction results at the next time can be obtained.Experiments show that the prediction accuracy of the algorithm is about80%.In general,this paper studies the scheme design,data acquisition,vehicle detection algorithm,fast denoising algorithm and trajectory prediction algorithm in the field of DAS expressway vehicle monitoring,and has achieved the considerable results.