GridSmart goes beyond detection. GridSmart is a new class of product known as three-dimensional omni-directional vehicle tracking.
Much like loops, regular multi-camera video vehicle detection operates much like a tripwire. Vehicles that pass over the loop or into a video detection zone trigger a simple on / off logic to make calls to the traffic controller. With regular multi-camera video vehicle detection, this tripwire logic is based on a premise called background subtraction. A background image is modeled within the video, and this background is subtracted from each image to determine differences. Where differences are detected, the tripwire is activated and a call is made.
GridSmart applies object tracking algorithms for more advanced functionality and enhanced performance.
Object Tracking
As an object (vehicle, bicycle, pedestrian) moves, GridSmart assigns tracking points to the edges of the object and follows those points throughout the intersection. Based on how those points move, independently and in relation to one another, a three-dimensional (3D) classification model (vehicle – length based, pedestrian, or other) is assigned to the object and given a unique tracking identification (ID) number for continued tracking throughout the camera’s horizon-to-horizon field of view. The model can change, while retaining the same unique ID number, as the confidence level improves with each frame and the vehicle, pedestrian, bicycle, or other object approaches. With GridSmart not simply acting as a “tripwire” detector, points of interest tracking in an unexpected direction will be ignored or flag an alert. Tracking also allows for objects to move “through” other objects and occlusions…much like the human eye, GridSmart knows that objects still exist when they are behind another object.
Three-dimensional Vehicle Modeling
As referenced above, vehicle classification models are three-dimensional. With GridSmart measuring in real-world dimensions, not pixels, these three dimensional models allow for perspective to be applied allowing for more accurate detection, vehicle classification, and average speed calculations. By incorporating the geographic coordinates of the camera, the GridSmart engine is also able to anticipate shadow placement and mitigate many of the issues that shadows creates with traditional video solutions.
