A Real-Time Street Actions Detection
Salah Alghyaline · International Journal of Advanced Computer Science and Applications · 2019
Human action detection in real time is one of the most important and challenging problems in computer vision. Nowadays, CCTV cameras exist everywhere in our lives. However, the contents of these cameras are monitored and analyzed using human operator. This paper proposes a real time human action detection approach which efficiently detects basic and common actions in the street such as stopping, walking, running, group stopping, group walking, and group running. The proposed approach measures the object movement type based on three techniques: YOLO object detection, Kalman Filter and Homography. Real videos from CCTV camera and BEHAVE dataset are used to test the proposed method. The experimental results show that the proposed method is very effective and accurate to detect basic human actions in the street. The accuracies of the proposed method on the tested videos are 96.9% and 88.4% for the BEHAVE and the created CCTV datasets, respectively. The proposed approach runs in real time with more than 50 fps for BEHAVE dataset and 32 fps for the created CCTV datasets.