As the storage time and space constraints, existing pedestrian detection algorithms cannot meet the real-time intelligent video surveillance. An approach based on motion analysis is proposed to solve this problem. The method only extracts gradient histogram features from the objects containing motion information. It avoids window scan on the whole image, and save the processing of Pyramid transformation. So the efficiency of image processing is greatly improved. Compared with conventional pedestrian detection, experimental results show that the approach is fast and effective, and can meet the real-time needs of intelligent video surveillance.