Traffic Incident Detection Research Articles

Learning Traffic as Images for Incident Detection Using Convolutional Neural Networks

The timely and accurate detection of traffic incidents is beneficial to reduce associated economic losses and avoid secondary crashes. Inspired by the impressive success of the image classification algorithms, especially convolutional neural networks (CNNs), this study proposes a novel framework to detect highway traffic incidents by learning the traffic state as images. In such a framework, the probe vehicles equipped with the global positioning system equipment are used to obtain data. The Gramian Angular Difference Fields and Piecewise Aggregation Approximation algorithms are used to convert the link speed time series data into images. CNNs can extract the traffic features based on these images and consider an incident detection problem as a binary classification task. Further, the effectiveness of the proposed framework is evaluated by applying it to detect the traffic in a real-world environment, i.e., the Guangzhou Airport Expressway. The results illustrate that the proposed model outperforms several other algorithms with respect to almost all the performance indexes, including the detection rate with different false alarm rates and the area under the receiver operating characteristic curve.

Traffic Incident Detection Method Based on Factor Analysis and Weighted Random Forest

Timely and accurate detection of traffic incidents can effectively reduce personal casualties and property losses, and improve the ability of macro-control and scientific decision-making of traffic. The unbalance of traffic incident data has a great influence on the detection effect. Therefore, a traffic incident detection method based on factor analysis and weighted random forest (FA-WRF) is designed. Through the analysis of the change rule of traffic flow parameters to build the initial incident variable. The factor analysis (FA) method is used to reduce the dimension of the initial incident variables. Using Bootstrap improved algorithm to predetermine the data extraction standard of the training set. The MCC coefficient value is calculated for the classification effect of the decision tree after training, and is assigned to each tree as a weight value, so as to ensure that the trees with better classification ability have more voting power in the voting process, thus improve the overall classification performance of the random forest (RF) algorithm for unbalanced data. The detection performance is evaluated by the common criteria including the detection rate, the false alarm rate, the classification rate and the area under the curve of the receiver operating characteristic (AUC). Based on the location detector data from expressway, the incident data in which accounts for 6.5%, showing a typical unbalance. The experimental results indicate that the model based on FA-WRF has the better classification effect. Meanwhile it is competitive in processing unbalanced data classification compared with Support Vector Machine.

Detection and Tracking of Vehicles Based on Colour Probability Density

Vehicle monitoring is a very important part in the intelligent transportation systems towards real-time monitoring of intersection traffic condition, the dynamic traffic incident detection and traffic parameter extraction. This paper proposes a vehicle tracking method based on mean shift. During the detection period, tracking objects of vehicles are constructed. The current vehicle position is predicted from the target area of former frame. In the candidate area of the target image, foreground area mask is adopted as a condition whether a pixel is selected; this makes the colour probability density to more accurately reflect the characteristics of the vehicle, and avoids the background region's influence on the mean shift iterations. Experiments show that this method can effectively detect the position of the vehicle, and provides an effective vehicle tracking method in the intelligent transportation system.

Metaheuristic approach to optimize placement of detectors in transport networks — case study of Serbia

Proper number and optimal location of detectors in transport networks enable early traffic incident detection and collecting other relevant traffic flow data. Increasing number of located detectors in networks provides accuracy of obtained data, while requiring more investments and maintenance cost support. The detectors need to be placed such that they can successfully sample the traffic conditions with the least possible error. On the other hand, traffic authorities tend to minimize the number of located detectors on the network to achieve investment savings. The proposed model provides optimal locations of a finite set of detectors on a highway corridor, taking into account minimization of travel time estimation error and constraints of available capital cost. The bee colony optimization metaheuristic based on the improvement concept is used to solve the formulated problem. The proposed algorithm is tested on a real data collected in the Republic of Serbia.

Sensing Urban Transportation Events from Multi-Channel Social Signals with the Word2vec Fusion Model.

Social sensors perceive the real world through social media and online web services, which have the advantages of low cost and large coverage over traditional physical sensors. In intelligent transportation researches, sensing and analyzing such social signals provide a new path to monitor, control and optimize transportation systems. However, current research is largely focused on using single channel online social signals to extract and sense traffic information. Clearly, sensing and exploiting multi-channel social signals could effectively provide deeper understanding of traffic incidents. In this paper, we utilize cross-platform online data, i.e., Sina Weibo and News, as multi-channel social signals, then we propose a word2vec-based event fusion (WBEF) model for sensing, detecting, representing, linking and fusing urban traffic incidents. Thus, each traffic incident can be comprehensively described from multiple aspects, and finally the whole picture of unban traffic events can be obtained and visualized. The proposed WBEF architecture was trained by about 1.15 million multi-channel online data from Qingdao (a coastal city in China), and the experiments show our method surpasses the baseline model, achieving an 88.1% F1 score in urban traffic incident detection. The model also demonstrates its effectiveness in the open scenario test.

Travel Time Functions Prediction for Time-Dependent Networks

The studies on the TDN (time-dependent network), in which the travel time of the same road segment varies depending on the time of the day, have attracted much attention of researchers, but there is little work focusing on the travel time functions prediction problem. Though traditional methods for travel time or travel speed prediction problem can be used to generate the travel time functions, they have some limitations due to the need of less breakpoints, fine granularity, and long-term prediction. In this paper, we study the travel time functions prediction problem for TDN based on taxi trajectory data. In order to maintain a high degree of accuracy in fine-grained and long-predicted situations, we take into account not only the traffic incidents but also the data sparsity. Specifically, a traffic incident detection method is proposed based on k-means algorithm and a downstream-based strategy is proposed to estimate the speeds of segments considering the data sparsity. To make the breakpoints of function not so much, a prediction algorithm based on classification using ELM (extreme learning machine) is proposed, which predicts the speed classes taking both the weather and the adjacent segment conditions into account. In addition, a transformation method is presented to convert the discrete travel speeds into piecewise linear functions satisfying FIFO (First-In-First-Out) property. The experimental results show that ELM outperforms SVM (support vector machine) with regard to both the training time and prediction accuracy. Moreover, it also can be seen that both the weather conditions and the adjacent segment conditions have impact on the prediction accuracy.

SVM and KNN ensemble learning for traffic incident detection

Traffic incident detection is a very important research area of intelligent transportation systems. Many methods have obtained good performance in traffic incident detection. However, the robustness of these methods is not satisfactory. Namely, when one method is applied on another data set again, its performance is not always good, even it had obtained good performance on one data set once. In this paper, we propose an ensemble learning method to improve the robustness in traffic incident detection. The proposed method trains individual SVM and KNN models firstly. And then, it takes a strategy to combine them for better final output. Experimental results show that the propose method has achieved the best performance among all the compared methods. Also, the ensemble learning strategy in the proposed method has improved the robustness of the individual models.

Text Based Approach For Similar Traffic Incident Detection from Twitter

Microblog has been used as an information source to detect real-world event. Several related studies retrieved road traffic event based on textual content. Not only detect traffic incident, we found that it is necessary to recognize statuses with similar traffic incident content. Better representation of traffic information will help the handling of traffic incident by related parties. This study proposes text-based approach for identification of similar traffic incident from twitter posts. The proposed approach performs traffic incident information extraction and calculates information’s weight based on textual similarity upon traffic incident information gained. We evaluate the proposed method by using a traffic incident information retrieval system. We used Indonesian language corpus contains traffic incident tweets data. Best average f-measure 70% was achieved by retrieval system that tested using Jaccard coefficient. Therefore text matching such as Jaccard coefficient is more suitable to be implemented in very short text document such as extracted tweet document. The experiment result gives the conclusion that the proposed approach can be implemented for identification of similar traffic incident information from Twitter. 

Detection of traffic incidents using nonlinear time series analysis.

In this study, we present results of the application of nonlinear time series analysis on traffic data for incident detection. More specifically, we analyze daily volume records of Attica Tollway (Greece) collected from sensors located at various locations. The analysis was performed using the Recurrence Plot (RP) and Recurrence Quantification Analysis (RQA) method of the volume data of the lane closest to the median. The results show that it is possible to identify, through the abrupt change of the dynamics of the system revealed by RPs and RQA, the occurrence of incidents on the freeway and differentiate from recurrent traffic congestion. The proposed methodology could be of interest for big data traffic analysis.

On fine-grained geolocalisation of tweets and real-time traffic incident detection

Recently, geolocalisation of tweets has become important for a wide range of real-time applications, including real-time event detection, topic detection or disaster and emergency analysis. However, the number of relevant geotagged tweets available to enable such tasks remains insufficient. To overcome this limitation, predicting the location of non-geotagged tweets, while challenging, can increase the sample of geotagged data and has consequences for a wide range of applications. In this paper, we propose a location inference method that utilises a ranking approach combined with a majority voting of tweets, where each vote is weighted based on evidence gathered from the ranking. Using geotagged tweets from two cities, Chicago and New York (USA), our experimental results demonstrate that our method (statistically) significantly outperforms state-of-the-art baselines in terms of accuracy and error distance, in both cities, with the cost of decreased coverage. Finally, we investigated the applicability of our method in a real-time scenario by means of a traffic incident detection task. Our analysis shows that our fine-grained geolocalisation method can overcome the limitations of geotagged tweets and precisely map incident-related tweets at the real location of the incident.

Comparing detection and disclosure of traffic incidents in social networks: an intelligent approach based on Twitter vs. Waze

Nowadays, social networks have become in a communication medium widely used to disseminate any type of information. In particular, the shared information in social networks usually includes a considerable number of traffic incidents reports of specific cities. In light of this, specialized social networks have emerged for detecting and disseminating traffic incidents, differentiating from generic social networks in which a wide variety of topics are communicated. In this context, Twitter is a case in point of a generic social network in which its users often share information about traffic incidents, while Waze is a social network specialized in traffic. In this paper we present a comparative study between Waze and an intelligent approach that detects traffic incidents by analyzing publications shared in Twitter. The comparative study was carried out considering Ciudad AutA³noma de Buenos Aires (CABA), Argentina, as the region of interest. The results of this work suggest that both social networks should be considered as complementary sources of information. This conclusion is based on the fact that the proportion of mutual detections, i.e. traffic incidents detected by both approaches, was considerably low since it did not exceed 6% of the cases. Moreover, the results do not show that any of the approaches tend to anticipate in time to the other one in the detection of traffic incidents.

Automatic Vehicle Accident Detection Based on GSM System

The evolution of technology has increased the rate of traffic accidents that occurs frequently causing loss of lifeand property. Therefore, the automatic traffic monitoring system gradually attracted the attention of researchers in improvingtraffic safety through the field of intelligent transport systems. In this paper, a cost-effective approach based on GSM systemfor automatic traffic incident detection is proposed. This paper provides an optimal solution to reduce the death rate by usingthe vibration sensor and GSM system; the implementation system is based on hardware (circuits) and software to build agraphical user interface (GUI) using LabView TM to process the data. Sensors are installed into the vehicles on each side ofthe vehicle. An SMS will be sent to the user after the accident. This system will assist in search and rescue vehicle involved inthe accident.

Fuzzy deep learning based urban traffic incident detection

Traffic incident detection (TID) is an important part of any modern traffic control because it offers an opportunity to maximise road system performance. For the complexity and the nonlinear characteristics of traffic incidents, this paper proposes a novel fuzzy deep learning based TID method which considers the spatial and temporal correlations of traffic flow inherently. Parameters of the deep network are initialized using a Stacked Auto-Encoder (SAE) model following a layer by layer pre-training procedure. To conduct the fine tuning step, the back-propagation algorithm is used to precisely adjust the parameters in the deep network. Fuzzy logic is employed to control the learning parameters where the objective is to reduce the possibility of overshooting during the learning process, increase the convergence speed and minimize the error. To find the best architecture of the deep network, we used a separate validation set to evaluate different architectures generated randomly based on the Mean Squared Error (MSE). Simulation results show that the proposed incident detection method has many advantages such as higher detection rate and lower false alarm rate.

Automatic Vehicle Accident Detection Based on GSM System

The evolution of technology has increased the rate of traffic accidents that occurs frequently causing loss of life and property. Therefore, the automatic traffic monitoring system gradually attracted the attention of researchers in improving traffic safety through the field of intelligent transport systems. In this paper, a cost-effective approach based on GSM system for automatic traffic incident detection is proposed. This paper provides an optimal solution to reduce the death rate by using the vibration sensor and GSM system; the implementation system is based on hardware (circuits) and software to build a graphical user interface (GUI) using LabView TM to process the data. Sensors are installed into the vehicles on each side of the vehicle. An SMS will be sent to the user after the accident. This system will assist in search and rescue vehicle involved in the accident.

Traffic Incident Detection Based on Extreme Machine Learning

ABSTRACT Traffic incident detection is an essentialpart of the intelligenttransportation systemand attracts lots of attention from researchers in different areas. To reduce the computation and to avoid overfitting of the traditional feed-forward neural network, an extreme machine learning is implemented to detect traffic incidents. Using the real-world data, the proposed model is compared with feed-forward neural network and other two commonly used models by several evaluation criteria. The results indicate the proposed model has the highest accuracy, detection rate and lowest false alarm rate. Moreover, it consumes less time. In conclusion, the performance of the proposed model is better than other benchmark models and suit to detect traffic incidents.

Traffic volume responsive incident detection

Abstract Incident detection is one of the major concerns of freeway operators, as incidents account for more than 60% of the travel delays faced by motorists, especially under recurrent traffic congestion. Algorithms detecting incidents are categorized based on the intelligence they use to analyse the measurements taken by traffic monitoring (volume, occupancy, speed). As roadway geometry and traffic conditions affect the algorithms’ accuracy, the objectives of this paper is to take them into account when specifying the threshold values, and to assess the effect on the algorithms’ performance on the network operations. The main steps of the methodology are the calibration of available incident detection algorithms (California #7 και DELOS) to a set of traffic and incident data and the validation of the threshold values. Further adaptation of these values is attempted to the changing traffic volumes. When the threshold values were being calibrated on the prevailing traffic conditions, an improvement was observed on all three performance indices (detection rate increase by 20%, false alarm rate decrease by 25% and time to detect at the same levels).

Automatic Incident Detection in Intelligent Transportation Systems Using Aggregation of Traffic Parameters Collected Through V2I Communications

Recent research in Intelligent Transportation System (ITS) focuses on Automatic Incident Detection (AID) techniques. Using advances in wireless networking and sensor technologies, modern vehicles have the ability to communicate with each other as well as with roadside infrastructure units (RSUs) in order to increase road safety. These new innovations in transportation technology provide traffic management system the ability to use data collected from the vehicles on the road to detect congestion and traffic incidents. Lately, many techniques were developed to alert drivers in advance about traffic incidents and to enable them to avoid congestion. In this paper, we discuss probabilistic collection of traffic data through vehicle-to-infrastructure (V2I) communications and present two novel techniques for automatic detection of traffic incidents in a highway scenario that are based on the use of distance and time for changing lanes, respectively vehicle speed changes over time. The proposed methods, which are illustrated with numerical results obtained from simulations, outperform alternative AID techniques through higher incident detection rates, about 25% shorter peak queue values and 20% faster dissipation of roadway congestion.

Incident Detection Through Twitter

Traffic incident information is disseminated via Twitter from various types of accounts. It is more common to find active or transitive verbs and adverbs in tweets from personal accounts because individuals report personal experiences (e.g., “just saw an accident”). Tweets from an organization or agency [e.g., 511 (a telephone hotline for transportation information widely used in the United States and Canada), departments of transportation] are more structured and commonly include nouns and past participles (e.g., “one lane blocked”). Organization accounts mostly provide incident location, type, severity, and so on, whereas personal tweets do not usually provide such details. However, an agency tweets about an incident usually after the incident management (IM) officials have already been notified. Because of this timing, a personal tweet is more likely to carry useful “new” information for IM purposes. This study investigated the detection of traffic incidents through Twitter feeds by using these differences in structure and information content found in organization and personal social media accounts. Tweets collected via the Twitter public application programming interface were manually coded and treated separately as either personal or organizational, and the “dictionaries” used to perform relevancy classification were derived separately. Combinations of dictionaries (i.e., personal only, organizational only, personal and organizational) were used for “term frequency–inverse document frequency” and naïve Bayesian analysis. It was shown that analysis specific to account types helped achieve better accuracy in classification for targeting relevant tweets. Therefore, account-specific analysis should be considered for more efficient and effective event detection for IM purposes.

Detection of traffic congestion and incidents from GPS trace analysis

This paper presents an expert system for detecting traffic congestion and incidents from real-time GPS data collected from GPS trackers or drivers’ smartphones. First, GPS traces are pre-processed and placed in the road map. Then, the system assigns to each road segment of the map a traffic state based on the speeds of the vehicles. Finally, it sends to the users traffic alerts based on a spatiotemporal analysis of the classified segments. Each traffic alert contains the affected area, a traffic state (e.g., incident, slowed traffic, blocked traffic), and the estimated velocity of vehicles in the area. The proposed system is intended to be a valuable support tool in traffic management for municipalities and citizens. The information produced by the system can be successfully employed to adopt actions for improving the city mobility, e.g., regulate vehicular traffic, or can be exploited by the users, who may spontaneously decide to modify their path in order to avoid the traffic jam. The elaboration performed by the expert system is independent of the context (urban o non-urban) and may be directly employed in several city road networks with almost no change of the system parameters, and without the need for a learning process or historical data. The experimental analysis was performed using a combination of simulated GPS data and real GPS data from the city of Pisa. The results on incidents show a detection rate of 91.6%, and an average detection time lower than 7min. Regarding congestion, we show how the system is able to recognize different levels of congestion depending on different road use.

Extreme Learning Machine-Based Traffic Incidents Detection with Domain Adaptation Transfer Learning

Abstract Traffic incidents in big cities are increasing alongside economic growth, causing traffic delays and deteriorating road safety conditions. Thus, developing a universal freeway automatic incident detection (AID) algorithm is a task that took the interest of researchers. This paper presents a novel automatic traffic incident detection method based on the extreme learning machine (ELM) algorithm. Furthermore, transfer learning has recently gained popularity as it can successfully generalise information across multiple tasks. This paper aimed to develop a new approach for the traffic domain-based domain adaptation. The ELM was used as a classifier for detection, and target domain adaptation transfer ELM (TELM-TDA) was used as a tool to transfer knowledge between environments to benefit from past experiences. The detection performance was evaluated by common criteria including detection rate, false alarm rate, and others. To prove the efficiency of the proposed method, a comparison was first made between back-propagation neural network and ELM; then, another comparison was made between ELM and TELM-TDA.