This dataset reports the progress of the Austin Transportation Department's Annual Signal Timing Program. Traffic signal engineers re-time approximately 1/3 of the city’s 1,000+ signals each year, with the goal of ensuring signals are timed for optimum safety and performance. This data powers our Signal Re-timing dashboard, available here: https://data.mobility.austin.gov/signal-timing You may also be interested in our dataset of traffic signals by re-timing corridor, available here: https://data.austintexas.gov/Transportation-and-Mobility/Synchronized-Traffic-Signal-Corridors/efct-8fs9

Problem Statement

👉 Download the case studies here

Urban areas worldwide face increasing traffic congestion due to rapid urbanization and rising vehicle density. A city’s transportation department struggled with inefficient traffic flow, leading to longer travel times, increased fuel consumption, and higher emissions. Traditional traffic management systems were reactive rather than predictive, requiring a smarter, data-driven solution to address these issues.

Challenge

Developing an intelligent traffic management system involved tackling several challenges:

Collecting and processing real-time traffic data from multiple sources, including sensors, cameras, and GPS devices.

Predicting traffic patterns and optimizing signal timings to reduce congestion.

Ensuring scalability to handle the growing urban population and vehicle density.

Solution Provided

An AI-powered traffic management system was developed using advanced algorithms, real-time data analytics, and IoT sensors. The solution was designed to:

Monitor and analyze traffic flow in real time using data from IoT-enabled sensors and connected vehicles.

Optimize traffic signal timings dynamically to minimize congestion at key intersections.

Provide actionable insights to city planners for long-term infrastructure improvements.

Development Steps

Data Collection

Installed IoT sensors at intersections and leveraged data from traffic cameras and connected vehicles to gather real-time traffic data.

Preprocessing

Cleaned and processed the collected data to identify patterns, peak congestion times, and traffic bottlenecks.

AI Model Development

Developed machine learning models to predict traffic flow and congestion based on historical and real-time data. Implemented optimization algorithms to adjust traffic signal timings dynamically.

Simulation & Validation

Tested the system in simulated environments to evaluate its effectiveness in reducing congestion and improving traffic flow.

Deployment

Deployed the system across key urban areas, integrating it with existing traffic control systems for seamless operation.

Continuous Monitoring & Improvement

Established a feedback loop to refine models and algorithms based on real-world performance and new traffic data.

Results

Decreased Traffic Congestion

The system reduced congestion by 25%, resulting in smoother traffic flow across the city.

Improved Travel Times

Optimized traffic management led to significant reductions in average travel times for commuters.

Enhanced Urban Mobility

Efficient traffic flow improved access to key areas, benefiting both residents and businesses.

Reduced Environmental Impact

Lower congestion levels minimized fuel consumption and greenhouse gas emissions, contributing to sustainability goals.

Scalable and Future-Ready

The system’s modular design allowed easy expansion to new areas and integration with emerging transportation technologies.

Data were collected during the Multi-Modal Intelligent Transportation Signal Systems (MMITSS) study. MMITSS is a next-generation traffic signal system that seeks to provide a comprehensive traffic information framework to service all modes of transportation. The GPS data set catalogs the vehicle operation data of the test vehicles that used for the MMITSS field testing. The data contains the performance and operation details of vehicles. This file contains a number of fields detailing elements such as vehicle position and speed, fidelity measures of GPS-based data elements, and vehicle operation data. NOTE: All extra attachments are located in Multi-Modal Intelligent Traffic Signal Systems Basic Safety Message

Overview

The original Udacity Self Driving Car Dataset is missing labels for thousands of pedestrians, bikers, cars, and traffic lights. This will result in poor model performance. When used in the context of self driving cars, this could even lead to human fatalities.

We re-labeled the dataset to correct errors and omissions. We have provided convenient downloads in many formats including VOC XML, COCO JSON, Tensorflow Object Detection TFRecords, and more.

Some examples of labels missing from the original dataset: https://i.imgur.com/A5J3qSt.jpg" alt="Examples of Missing Labels">

Stats

The dataset contains 97,942 labels across 11 classes and 15,000 images. There are 1,720 null examples (images with no labels).

All images are 1920x1200 (download size ~3.1 GB). We have also provided a version downsampled to 512x512 (download size ~580 MB) that is suitable for most common machine learning models (including YOLO v3, Mask R-CNN, SSD, and mobilenet).

Annotations have been hand-checked for accuracy by Roboflow.

https://i.imgur.com/bOFkueI.pnghttps://" alt="Class Balance">

Annotation Distribution: https://i.imgur.com/NwcrQKK.png" alt="Annotation Heatmap">

Use Cases

Udacity is building an open source self driving car! You might also try using this dataset to do person-detection and tracking.

Using this Dataset

Our updates to the dataset are released under the MIT License (the same license as the original annotations and images).

Note: the dataset contains many duplicated bounding boxes for the same subject which we have not corrected. You will probably want to filter them by taking the IOU for classes that are 100% overlapping or it could affect your model performance (expecially in stoplight detection which seems to suffer from an especially severe case of duplicated bounding boxes).

About Roboflow

Roboflow makes managing, preprocessing, augmenting, and versioning datasets for computer vision seamless.

Developers reduce 50% of their boilerplate code when using Roboflow's workflow, save training time, and increase model reproducibility. :fa-spacer:

Here are a few use cases for this project:

1. Autonomous Vehicle Navigation: FFDR V1 could be used to identify, classify, and interpret various objects on the road for self-driving car development. This application can help autonomous vehicles decipher different elements such as cars, traffic lights, stop signs, and pedestrians in real-time, thereby facilitating safe and efficient navigation.

2. Traffic Management Systems: The model can assist in monitoring and analyzing traffic flow in urban areas, helping authorities make data-driven decision-making for traffic light timing or creating more efficient traffic routes.

3. Surveillance & Public Safety: FFDR V1 can be deployed in surveillance systems to identify vehicles, analyze traffic patterns, and detect anomalous behavior for crime prevention or investigative purposes. It can also detect pedestrians and cyclists in potential collision scenarios improving public safety.

4. Augmented Reality (AR) Games: The model can be utilized for creating hyper-realistic AR games that interact with real-world vehicles, bikes, and pedestrians, offering an enhancing gaming experience for users.

5. Intelligent Transportation Systems: FFDR V1 can serve in optimizing public transport systems by helping count and classify vehicles like buses, trains, motorbikes, etc., thereby aiding in transit planning, route optimization, and performance evaluation.

This dataset will no longer be updated. Selected key performance indicators on the TFL Road Network. Number of hours of Serious and Severe Disruption on the road network by planned and unplanned status, journey time reliability, and total number of works undertaken on the road network. TLRN = TFL Road Network The maximum permissibile total number of road works allowed on the TLRN was capped at 3,250 for any one period, in Period 1 of 2013/14 until the end of the financial year 2014/15. This is a reduction of 13.4 per cent from the cap of 3,753 that applied from Period 7 2011/12 to the end of financial year 2012/13. The key measure for monitoring smoothing traffic flow is journey time reliability. It is defined as the percentage of journeys completed within an allowable excess of 5 minutes for a standard 30 minute journey during the AM peak. Serious congestion. There is traffic congestion that is unusual for the time of day at the location or in an area and traffic has been stopped for less than 5 minutes - but in excess of the red signal time displayed on the traffic signals operating on the road. Severe congestion. There is traffic congestion that is unusual for the time of day at the location or in an area and traffic has been stopped for more than 5 minutes; and traffic queuing is longer than normal for the time of day, more than for ‘serious’ congestion. TFL also publish annual reports on road safety here.

This data contains information about traffic and pedestrian signals in the city of Austin, Texas. The data is updated on a daily basis and maintained by the Austin Transportation & Public Works Department's Arterial Management Division.

You may also be interested in the following resources:

• Traffic Signal Monitor https://data.mobility.austin.gov/signal-monitor

• Signal Evaluations https://data.mobility.austin.gov/signal-evaluations

City of Topeka traffic poles. This dataset consist of traffic signal poles, warning flashers and pedestrian crossing poles.

This dataset presents the IoT network traffic generated by connected objects. In order to understand and characterise the legitimate behaviour of network traffic, a platform is created to generate IoT traffic under realistic conditions. This platform contains different IoT devices: voice assistants, smart cameras, connected printers, connected light bulbs, motion sensors, etc. Then, a set of interactions with these objects is performed to allow the generation of real traffic. This data is used to identify anomalies and intrusions using machine learning algorithms and to improve existing detection models. Our dataset is available in two formats: pcap and csv and was created as part of the EU CEF VARIoT project https://variot.eu. To download the data in pcap format and for more information, our database is available on this web portal : https://www.variot.telecom-sudparis.eu/.

Motivation

Drowsiness is an intermediate condition that fluctuates between alertness and sleep. It reduces the consciousness level andhinders a person from responding quickly to important road safety issues [1]. The American Automobile Association (AAA) has reported that about 24% of 2,714 drivers that participated in a survey revealed being extremely drowsy while driving, at least once in the last month [2]. In 2017, the National Highway Transportation Safety Administration (NHTSA) also reported 795 fatalities in motor vehicle crashes involving drowsy drivers [3]. Drowsy driving has caused about 2.5% of fatal accidents from 2011 through 2015 in the USA, and it is estimated to produce an economic loss of USD 230 billion annually [4]. Klauer et al. have found in their study that drowsy drivers contributed to 22-24% of crashes or near-crash risks [5]. The German Road Safety Council (DVR) has reported that one out of four fatal highway crashes has been caused by drowsy drivers [6]. In a study carried out in 2015, it has been reported that the average prevalence of falling asleep while driving in the previous two years was about 17% in 19 European countries [6]. The results of these studies emphasize the importance of detecting drowsiness early enough to initiate preventive measures. Drowsiness detection systems are intended to warn the drivers before an upcoming level of drowsiness gets critical to prevent drowsiness-related accidents.

Intelligent Systems that automate motor vehicle driving on the roads are being introduced to the market step-wise. The Society of Automotive Engineers (SAE) issued a standard defining six levels ranging from no driving automation (level 0) to full driving automation (level 5) [7]. While the SAE levels 0-2 require that an attentive driver carries out or at least monitors the dynamic driving task, in the SAE level 3 of automated driving, drivers will be allowed to do a secondary task allowing the system to control the vehicle under limited conditions, e.g., on a motorway. Still, the automation system has to hand back the vehicle guidance to the driver whenever it cannot control the state of the vehicle any more. However, the handover of vehicle control to a drowsy driver is not safe. Therefore, the system should be informed about the state of the driver.

To date, different Advanced Driver Assistance Systems (ADAS) have been made by car manufactures and researchers to improve driving safety and manage the traffic flow. ADAS systems have been benefited from advanced machine perception methods, improved computing hardware systems, and intelligent vehicle control algorithms. By recently increasing the availability of huge amounts of sensor data to ADAS, data-driven approaches are extensively exploited to enhance their performance. The driver drowsiness detection systems have gained much attention from researchers. Before its use in the development of driving automation, drowsiness warning systems have been produced for the direct benefit of avoiding accidents.

The aim of the WACHSens project was to collect a big data set to detect the different levels of driver drowsiness during performing two different driving modes: manual and automated.

To retrieve this data set, please send a request to: arno.eichberger@tugraz.at

References:

[1] M. Awais, N. Badruddin, and M. Drieberg, "A Hybrid Approach to Detect Driver Drowsiness Utilizing Physiological Signals to Improve System Performance and Wearability,"Sensors, vol. 17, no. 9, 2017, doi: 10.3390/s17091991

[2] AAA Foundation for Traffic Safety, "2019 Traffic Safety Culture Index (Technical Report), June 2020," Washington, D.C., Jun. 2020. [Online]. Available: https://aaafoundation.org/2019-traffic-safety-culture-index/

[3] National Highway Traffic Safety Administration, "Traffic Safety Facts: 2017 Fatal Motor Vehicle Crashes: Overview," NHTSA's National Center for Statistics and Analysis, 1200 New Jersey Avenue SE., Washington DOT HS 812 603, Oct. 2018. Accessed: Apr. 14 2021. [Online]. Available: https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/812603

[4] Agustina Garcés Correa, Lorena Orosco, and Eric Laciar, "Automatic detection of drowsiness in EEG records based on multimodal analysis," Medical Engineering & Physics, vol. 36, no. 2, pp. 244–249, 2014, doi: 10.1016/j.medengphy.2013.07.011

[5] S. Klauer, V. Neale, T. Dingus, Jeremy Sudweeks, and D. J. Ramsey, "The Prevalence of Driver Fatigue in an Urban Driving Environment : Results from the 100-Car Naturalistic Driving Study," in 2006.

[6] Fraunhofer-Gesellschaft,Eyetracker warns against momentary driver drowsiness - Press Release Oktober 12, 2010. [Online]. Available: https://www.fraunhofer.de/en/press/research-news/2010/10/eye-tracker-driver-drowsiness.html (accessed: Apr. 14 2021).

[7] T. Inagaki and T. B. Sheridan, "A critique of the SAE conditional driving automation definition, and analyses of options for improvement," Cogn Tech Work, vol. 21, no. 4, pp. 569–578, 2019, doi: 10.1007/s10111-018-0471-5

Performance comparison with popular detection models on the CCTSDB2021 dataset.

This dataset presents SUMO simulation results from a study employing a Proximal Policy Optimization (PPO) reinforcement learning model for dynamic traffic signal control at a single intersection. The simulation, set in a mixed traffic environment, integrates diverse vehicle types and realistic traffic patterns to train an intelligent controller. The PPO model utilizes state information such as vehicle counts, waiting times, and signal phases, optimizing the traffic flow by reducing waiting times, minimizing queue lengths, and increasing throughput. This concise output offers valuable insights for researchers exploring adaptive traffic management and intelligent transportation systems in urban environments with promising performance results.