A method for robust recognition and tracking of multiple objects

Abstract

This paper presents an accurate and flexible method for robust recognition and tracking of multiple objects in video sequence. We calculate color moments and wavelet moments for each detected object. Based on the extracted moment features, the SVM achieves optimal object recognition performance. The object recognition rate is above 98.53%. Since the tracking accuracy of feature matching method could be degraded by occlusion, we add a Kalman filter tracking framework based on object recognition to improve multiple objects tracking. The previous object recognition module improves the performance and the accuracy of the Kalman filter tracking framework. Results obtained suggest that our tracking algorithm is very effective and robust even in challenging tracking conditions like occlusion and background clutter.