scorpeeon
commented
10 years ago Describes a system for real-time traffic surveillance that includes moving object segmentation, background updating, feature extraction, vehicle tracking and classification. Has a great summary of the popular methods for traffic monitoring - that seems to be the most valuable part for us.
Moving object segmentation is first used to extract the contour of vehicles. Based on the contours of vehicles and their corresponding minimal bounding box, salient discriminative features of vehicles are obtained. The tracking of targets is then achieved by comparing these features and by measuring the minimum distance between two consecutive images.
Uses #146 for background model/subtraction. Does not actually use feature points, it uses other kinds of features, mostly about boundingbox: such as compactness, aspect ratio, area ratio. The tracking method itself looks relatively simple. The classification is only done into 2 categories: cars or bikes.
The testing was only done in a few (seemingly easy) videos, with not many cars. The measured accuracy for classification was 96.4%, 92.7% for cars and bikes. The average counting accuracy for the three situations was 96.9%.
FeatureBasedVehicleFlowAnalysisAndMeasurementForARealTimeTrafficSurveillanceSystem.pdf