STRIDE Project: Fly-By Image Processing for Real Time Congestion Mitigation (Project H2)

Traffic monitoring is the centerpiece of congestion mitigation and traffic management. Whilst surveillance technologies have matured enough to provide informative depiction for the traffic, the current state of the art systems cannot support immediate congestion problems. Proactive congestion mitigation requires a) real-time surveillance for traffic parameters, b) prediction for imminent congestion onset, in order to c) inform responsible parties to take immediate actions to prevent congestion. This framework is founded on short time analysis (1-5 minutes) which is not valid up to date. We foresee that using a “flock” of interconnected, self-managed drones, can establish a deployable system to perform immediate monitoring/assessment for traffic condition to infer if congestion is approached. The drones will use their own computational and communication capabilities to host an integrated reconnaissance platform that performs traffic monitoring and traffic analysis in real-time fashion.

Towards above long-term goal, instead of using Convolutional Neural Network (CNN) to detect vehicles, a faster technique called YOLOv3 is used. In this technique, a single neural network is used to the full image which divides the image into regions and predicts bounding boxes and probabilities for each region, then these bounding boxes are weighted by the predicted probabilities. This technique requires huge computational power and therefore, GPUs are used to process the videos recorded by drones’ cameras. By calibrating the camera using real values compared to their apparent values in images, the detected vehicles can be tracked. The targeted feature (herein, features correlated to traffic congestion) were reproduced utilizing a traffic simulation models. The proposed methodology was tested by collecting and investigating video images from drone.

The project if continued further, has the potential to advance the state of proactive traffic and congestion management by embedding a distributed, simulation-based traffic state prediction system within the integrated drone surveillance software to enable congestion mitigations actions to be undertaken before congestion happens rather than after traffic flow has already broken down.