Closed scorpeeon closed 10 years ago
According to #54: "simple method based on template matching and temporal consistency via object classification and motion detection. Their method can deal with partial occlusions. Template matching plays a similar role to appearance-based models but does not take into account the variable appearance of the object due to lighting changes, self-occlusions, and other complex 3-dimensional projection effects."
Rather old paper about an end-to-end method for detecting, classifying (based on the target's shape), and tracking moving targets. Classification to 3 category: human, vehicle or background clutter. Uses a simple method based on a combination of temporal differencing and image template matching, which is claimed to have a good performance.
In temporal differencing video frames separated by a constant time are compared to find regions which have changed. In template matching each video image is scanned for the region which best correlates to an image template. Template matching is good for stationary targets, temporal differencing is good for moving targets.
They measured >~80% precision for classification and it seemed to be able to track single vehicles for multiple minutes.
Simple, efficient method, might worth checking out.
MovingTargetClassificationAndTrackingFromRealtimeVideo.pdf