Georgia Leveraged the Use of Existing Traffic Cameras at Intersections to Monitor and Assess Traffic Safety Using Live Video Images.
This study explored the art of deep learning, multiple-object detection and tracking, and performed testing in the domain of traffic conflict monitoring and assessment. The study used the existing traffic cameras installed at major intersections to develop and deploy smart algorithms to scan live images and extract traffic conflicts in real time. For this purpose, an Artificial Intelligence (AI) enhanced computational system was developed to automate the detection and quantification of traffic conflict events as they occur in real time using traffic cameras currently installed by transportation agencies. Tests on simulated and actual video images demonstrated the promise of the proposed approach for detecting and quantifying traffic conflict events.
- Install traffic cameras above and close to intersection centers. The top-down camera view improves system detection and tracking accuracy.
- Reduce localization noises inherited from the image processing model due to the perspective view of cameras. Localization noises might result in false detection, although the extracted trajectories could be smoothed out to reduce such a risk.
- Allocate adequate computing resources for image reading and processing. This prevents computational bottlenecks during video image reading and the object detection inference and enables the system to process live images and analyze them in real time to detect, quantify and log traffic conflict events.
- Allocate adequate computation resources in proportion to the number of video sources to be processed in real time. If the algorithm is meant to operate on a central server that simultaneously processes multiple streaming video sources, then adequate computation resources must be allocated accordingly.
- Explore the option of a dual camera setting or a 3D camera. This will account for varying grades of intersections to improve accuracy of localization.