search

kundralaci / JAMResearch

0 stars 0 forks source link

Vehicle Detection Method using Haar-like Feature on Real Time System #70

Closed scorpeeon closed 10 years ago

scorpeeon commented 10 years ago

VehicleDetectionMethodUsingHaarLikeFeatureOnRealTimeSystem.pdf

scorpeeon commented 10 years ago

This paper used cameras placed on front of cars. So it always sees other cars' rears (which is rarely the case in our system) and has assumptions like symmetry. Haar-like features are used to get strong edge features and remove shadows.

Despite using Haar-like feafures, a database is not used in here. The method looks for features only appearing in the region of vehicles: their shadow and their vertical edges on both their sides. Then these candidate regions are checked if they are indeed vehicles based on their symmetry.

They achieved a detection rate of 91.2% for vehicles in the same lane, 76.2% for vehicles in the side lane (not appearing symmetrical enough). Because of the shadow detection, this method is sensitive to light conditions.