Pedestrian and Vehicle Classification Surveillance System for Street-Crossing Safety

This paper presents a framework for automatic pedestrian and motor vehicle classification in a streetcrossing safety surveillance system. The proposed method is a coarse to fine classification approach divided into two branches. The moving objects are detected, tracked and clustered into fast moving and slow moving categories according to a motion speed estimation method. This approach is applicable to identify motor vehicles and further differentiate cars from scooters using width and shape features. The second part identifies slow moving objects. We improved the Recurrent Motion Image (RMI) algorithm to sort out the pedestrians due to their high RMI value. Haar-like features and the Adaboost algorithm are then employed to distinguish between pedestrians and scooters. Cars and scooters are identified using the object aspect ratio (AR) and area feature. The experimental results show that the recognition rate for 320 objects achieved 92.5%. The proposed system is promising for application to the traffic monitoring surveillance system.