Closed scorpeeon closed 10 years ago
Detects two headlights of the vehicles then verifies candidates with a decision tree composed of feature-based and appearance-based classifiers. Claims to be stable and have good results even in bad conditions like distracting public lights. Also tracks the cars, can handle occlusions (with Kalman-filter). Camera is calibrated to have a projection between the camera's and the world coordinates. Considers scene geometry to look for the headlights at the right places and in the right sizes in the 3D space (on the monitored lanes, at the average height of headlights). It also does vehicle classification based on the width of the headlights (~1m: car, >1.5m: heavy vehicle).
NightTimeTrafficSurveillanceARobustFrameworkForMultiVehicleDetectionClassificationAndTracking.pdf