Use On-Board Equipment With Computer Vision and Machine Learning To Identify Work Zones and Relay Information on Roadwork Conditions and Location in Real-Time Using Connected Vehicle Networks.
Roadwork Images from 19 Cities in the United States Were Processed to Help Machine Learning Methods Identify Work Zone Information.
Pittsburgh, Pennsylvania, United States
Automatic Detection and Localization of Roadwork
Summary Information
Transportation agencies often use predictive analysis tools to deploy more efficient roadwork configurations to minimize the disruption of work zones to traffic. This study developed computer vision and machine learning methods for automatic detection and localization of work zones, so that the calculated information could then be shared with other drivers, enabling dynamic route planning for navigation systems, driver assist systems, and self-driving cars. The researchers adopted an approach in which objects commonly located within work zones were detected, and based on their proximity to each other, location relative to surfaces such as roads, sidewalks, or bike lanes would indicate whether or not a work zone was present. The researchers created a comprehensive roadwork dataset containing 8,556 images from 19 different U.S. cities, obtained from cameras mounted on vehicles driving around select cities to help with training the machine learning algorithm.
- Wirelessly Relay Roadwork Zone Information Using Computer Vision and Machine Learning to Identify and Share Work Zone Locations with Connected Vehicles. This technology can significantly enhance driver awareness, driver-assist systems, and automated vehicle navigation by providing real-time information about roadwork zones. Additionally, investing in infrastructure that wirelessly relays this information can benefit all drivers, including those without detection systems, by sharing work zone locations with connected vehicles.
- Actively capture imaging data in roadwork zones. Developing accurate computer vision models for automatic roadwork detection requires a comprehensive dataset with hundreds of thousands of images from varied locations, weather, lighting, and roadwork appearances. Equipping roadwork, public works, and service vehicles with cameras can help gather this essential data.
- Establish a reliable ground truth for roadwork dataset images. It is recommended to sponsor grants or challenges for manually annotating roadwork dataset images. While a unified detector shows promise, reliable model training requires labor-intensive and costly manual annotations, crucial for improving the accuracy of computer vision models in detecting roadwork zones.
