Background
This Presidential Discretion Internal Research (PDIR) program applied computer vision approaches to a broad range of transportation-focused incident detection challenges, relying only on inputs from a generic infrastructure-mounted traffic camera. Furthermore, the way these techniques were applied was done in such a way that the resulting detection algorithm can be used in a broad range of deployment environments with unique limitations, including embedded, cloud, and rack-mounted server deployments. Many state Department of Transportations (DOT) have limited budgets, restricting their field hardware deployments to digital highway signs and cameras. If these commodity roadway cameras, which are already deployed, can serve dual purposes as both a means for manual roadway observation by operators and a cost-effective, wide-spread incident sensing network, this will provide significant additional value to these DOTs and strengthen Division 10’s value proposition to prospective and existing customers.
Approach
Our approach leveraged the existing traffic camera infrastructure owned by these agencies to detect actionable data. The suite of proposed detection capabilities target vehicle-related occurrences of note to transportation agencies, including wrong way drivers, stalled vehicles, vehicle speed, and more. For all proposed capabilities, there is a common functional need to first be able to extract the individual objects present in the field of view, then track movement of those same objects across multiple video frames. Once objects within the video feed could be detected and tracked, higher level algorithms were built to derive actionable data these agencies can use to inform both internally and the traveling public.
Accomplishments
With our best object detection model and object tracking algorithm we have accomplished the following capabilities:
- Wrong Way Driver Detection – enable rapid emergency response.
- Vehicle Speed Detection – enable developing travel time messages.
- Vehicle Presence Detection – enable dynamically timing signalized intersections.
- Vehicle Counts Detection – enable quantifying road use to inform maintenance schedule.
- Stalled Vehicle Detection – enable identifying where to deploy roadside assistance.
- Dynamic Pan/Tilt/Zoom Road Tracking – enable detection with arbitrary field of view.
- Diverse Hardware Deployments.