Moving Pedestrian Localization and Detection

In this research, we collaborated with Universitas Syiah Kuala from Aceh to develop a deep learning model for detecting moving pedestrians from video footage. This model can be beneficial for intelligent surveillance systems, which is the main goal of the research. Generally, such systems are hindered by illuminations, cluttered backgrounds, and the size of the pedestrians. By using a robust background subtraction algorithm and YOLO, a deep learning model for this task was developed. 

The proposed model had proven to be both accurate and efficient. With a low prediction latency, it is considered suitable to be deployed in IoT for smart surveillance systems with limited computing power. The proposed algorithm can further be improved in the future by modifying the image pre-processing aspects to further improve the accuracy.