An intelligent transportation system (ITS) plays an important role in public transport management, security and other issues. Traffic flow detection is an important part of the ITS. Based on the real-time acquisition of urban road traffic flow information, an ITS provides intelligent guidance for relieving traffic jams and reducing environmental pollution. The traffic flow detection in an ITS usually adopts the cloud computing mode. The edge of the network will transmit all the captured video to the cloud computing center. However, the increasing traffic monitoring has brought great challenges to the storage, communication and processing of traditional transportation systems based on cloud computing. To address this issue, a traffic flow detection scheme based on deep learning on the edge node is proposed in this article. First, we propose a vehicle detection algorithm based on the YOLOv3 (You Only Look Once) model trained with a great volume of traffic data. We pruned the model to ensure its efficiency on the edge equipment. After that, the DeepSORT (Deep Simple Online and Realtime Tracking) algorithm is optimized by retraining the feature extractor for multiobject vehicle tracking. Then, we propose a real-time vehicle tracking counter for vehicles that combines the vehicle detection and vehicle tracking algorithms to realize the detection of traffic flow. Finally, the vehicle detection network and multiple-object tracking network are migrated and deployed on the edge device Jetson TX2 platform, and we verify the correctness and efficiency of our framework. The test results indicate that our model can efficiently detect the traffic flow with an average processing speed of 37.9 FPS (frames per second) and an average accuracy of 92.0% on the edge device.

An Edge Traffic Flow Detection Scheme Based on Deep Learning in an Intelligent Transportation System

Traffic forecasting is an important research area in Intelligent Transportation Systems that is focused on anticipating traffic in order to mitigate congestion. In this work we propose a deep neural network that simultaneously extracts the spatial features of traffic, using graph convolution, and its temporal features by means of Long Short Term Memory (LSTM) cells to make both short-term and long-term predictions. The model is trained and tested using sparse trajectory (GPS) data coming from the ride-hailing service of DiDi in the cities of Xi'an and Chengdu in China. Besides, presenting the deep neural network, we also propose a data-reduction technique based on temporal correlation to select the most relevant road links to be used as input. Combining the suggested approaches, our model obtains better results compared to high-performance algorithms for traffic forecasting, such as LSTM or the algorithms presented in the TRANSFOR19 forecasting competition. The model is capable of maintaining its performance over different time-horizons from 5 min to up to 4 h with multi-step predictions.

A graph CNN-LSTM neural network for short and long-term traffic forecasting based on trajectory data

PM2.5 is one of the most important pollutants related to air quality, and the increase of its concentration will aggravate the threat to people’s health. Therefore, the prediction of surface PM2.5 concentration is of great significance to human health protection. In this study, A hybrid CNN-LSTM model is developed by combining the convolutional neural network (CNN) with the long short-term memory (LSTM) neural network for forecasting the next 24h PM2.5 concentration in Beijing, which makes full use of their advantages that CNN can effectively extract the features related to air quality and the LSTM can reflect the long term historical process of input time series data. The air quality data of the last 7days and the PM2.5 concentration of the next day are first set as the input and output of the model due to the periodicity, respectively. Subsequently four models namely univariate LSTM model, multivariate LSTM model, univariate CNN-LSTM model and multivariate CNN-LSTM model, are established for PM2.5 concentration prediction. Finally, mean absolute error (MAE) and root mean square error (RMSE) are employed to evaluate the performance of these models and results show that the proposed multivariate CNN-LSTM model performs the best results due to low error and short training time.

/pdf/a-hybrid-cnn-lstm-model-for-forecasting-particulate-matter-4i5tbflx79.pdf

A Hybrid CNN-LSTM Model for Forecasting Particulate Matter (PM2.5)

Deep learning models for traffic flow prediction in autonomous vehicles: A review, solutions, and challenges

Urban flow prediction from spatiotemporal data using machine learning: A survey

The urban traffic flow prediction is a significant issue in the intelligent transportation system. In consideration of nonlinear and spatial-temporal features of urban traffic data, we propose a deep hybrid neural network improved by greedy algorithm for urban traffic flow prediction with taxi GPS trace. The proposed deep neural network model first combines the convolutional neural network (CNN), which extracts the spatial features, with the long short term memory (LSTM), which captures the temporal information, to predict urban traffic flow. Then, the proposed model is trained by a greedy policy to short time consumption and improves accuracy when a network goes deeper. Experimental results with real taxis GPS trajectory data from ${Xi'an}$ city show that the improved deep hybrid CNN-LSTM model can achieve higher prediction accuracy and shorter time consumption compared with existing methods.

Improved Deep Hybrid Networks for Urban Traffic Flow Prediction Using Trajectory Data

Predicting city-scale taxi origin-destination (OD) flows takes an important role in understanding passengers' travel demands as well as managing taxi operation and scheduling. But the complex spatial dependencies and temporal dynamics make this problem challenging. In this paper, a hybrid deep neural network prediction model based on convolutional LSTM (ConvLSTM) is proposed. For improving the prediction accuracy, the implicit correlation between travel time and OD flow is explored and they are combined as inputs of the prediction model. Moreover, in order to realize OD flows prediction at the road network level, and solve the problem that grid-based representation method cannot distinguish traffic flow at different heights, such as in multi-layer overpass areas, this paper presents a grid and road nested method to represent ODs. With the time of day partitioned into time slots, OD flows are extracted and predicted in both spatial and temporal domain. In the experiment, real taxi data are used to verify the proposed model and prediction method fully. And the experimental results show that the proposed model can effectively predict city-wide taxi OD flow, and outperforms the typical time sequence models and existing deep neural network models.

/pdf/prediction-of-city-scale-dynamic-taxi-origin-destination-4gxumcrqyf.pdf

Prediction of City-Scale Dynamic Taxi Origin-Destination Flows Using a Hybrid Deep Neural Network Combined With Travel Time

The invention discloses a method for improving map matching abnormal points based on a machine learning algorithm. The method comprises the following steps: acquiring projection coordinate point dataafter matching of one or more taxies as sample data, and then screening out normal matching projection points and abnormal matching projection points therefrom; for each abnormal matching projection coordinate point, screening out a candidate data set of the abnormal points from a correct matching data set; calculating the time similarity and the space similarity between each abnormal point and the projection points in the candidate data set of the abnormal points to obtain a space-time similarity set, and then calculating the mean value of space-time similarities according to the time similarity and the space similarity; comparing the space-time similarities with the mean value of the space-time similarities, acquiring the number of the time-space similarities greater than the mean valueof the space-time similarities, and using data meeting the condition as a final candidate data set; in the final candidate data set, calculating by a knn algorithm to obtain improved projection coordinate points. The method has the advantages of lower calculation amount and higher accuracy rate.

Method for improving map matching abnormal points based on machine learning algorithm

The invention discloses a vehicle OD flow prediction model construction method and a vehicle OD flow prediction method. In the vehicle OD flow prediction model construction method, a multi-granularityspace partitioning method using grid and road segment nesting is used for representing vehicle OD data at regional and road node levels; the number of trips and travel time between the OD are extracted simultaneously; and a deep prediction model LSTM-traf-deepCNN mixed with CNN and LSTM is used for predicting the OD flow combined with OD travel time. Compared with the traditional OD flow prediction method, the method disclosed by the invention fully considers the implicit relationship between the travel time and the OD flow, and trains the deep network by combining the travel time and the number of trips between the OD; and the obtained model has more accurate predictive power. The invention belongs to the technical field of transportation information engineering and can be used for predicting the OD flow of city-scale taxis.

Vehicle OD flow prediction model construction method and vehicle OD flow prediction method

Accurate urban traffic speed prediction plays a crucial role in intelligent transportation systems (ITSs). However, the potential spatial-temporal information of big urban traffic data captured from the complex behaviors of urban systems cannot be efficiently mined based on the state-of-art intelligent models. Therefore, we propose a hybrid Long Short-Term Memory (LSTM) and Restricted Boltzmann Machine (RBM) neural network with a fine-tuning strategy for urban traffic speed prediction. Firstly, the LSTM-RBM model dynamically combines the LSTM and RBM, such that the LSTM extracts the time-varying characteristics of the speed sequences to dynamically adjust the RBM, and then the RBM can capture the deep and detailed features of the speed sequences in the optimal way. Secondly, a transfer learning fine-tuning strategy is proposed to effectively pre-train the LSTM-RBM to achieve higher accuracy. Experimental results based on traffic speed data of the second ring road in Xi'an indicate that the proposed hybrid LSTM-RBM model with fine-tuning can outperform the existing deep models.

Ni Yuanyuan

Papers

Improved Deep Hybrid Networks for Urban Traffic Flow Prediction Using Trajectory Data

Prediction of City-Scale Dynamic Taxi Origin-Destination Flows Using a Hybrid Deep Neural Network Combined With Travel Time

Method for improving map matching abnormal points based on machine learning algorithm

Vehicle OD flow prediction model construction method and vehicle OD flow prediction method

Fine-Tuning Deep Hybrid Long Short-Term Memory and Restricted Boltzmann Machine Network for Urban Traffic Speed Prediction