search

kundralaci / JAMResearch

0 stars 0 forks source link

A real-time optical airborne road traffic monitoring system #51

Closed scorpeeon closed 10 years ago

scorpeeon commented 10 years ago

ARealtimeOpticalAirborneRoadTrafficMonitoringSystem.pdf

scorpeeon commented 10 years ago

This is an airborne system, it is installed to airplanes, the roads are observed from 500-3000 m height. The images are orthorectified, the roads are extracted (from road databases) and the cars are detected using Haar-like features. The paper's significant part is about the usage of Haar-like features for the car detection. The detector was trained with 2000 positive samples (in different lighting conditions).

It achieved a detection rate of around 90% (96% on one scene, 88% on another) and correctness rate of around 98%. Because of the placement of cameras, occlusions should not a problem here. The paper concludes that the images used in the training affects highly the performance of the detector. So for a good detector, samples from multiple areas, under different lighting conditions have to be used for the training.