Fast Human Detection for Video Event Recognition

Human body detection, which has become a research hotspot during the last two years, can be used in many video content analysis applications. This paper investigates a fast human detection method for volume based video event detection. Compared with other object detection systems, human body detection brings more challenge due to threshold problems coming from a wide range of dynamic properties. Motivated by approaches successfully introduced in facial recognition applications, it adapts and adopts feature extraction and machine learning mechanism to classify certain areas from video frames. This method starts from the extraction of Haar-like features from large numbers of sample images for well-regulated feature distribution and is followed by AdaBoost learning and detection algorithm for pattern classification. Experiment on the classifier proves the Haar-like feature based machine learning mechanism can provide a fast and steady result for human body detection and can be further applied to reduce negative aspects in human modelling and analysis for volume based event detection.