SpeedyGuard: Efficient abnormal behavior recognition for children via motion-triggered YOLOv4-tiny
Year: 2024
Type: conference-paper
Abstract: To stop children being alone at home from reaching assistance requirement that delays medical attendance-this paper proposes a new system called SpeedyGuard using a motion-triggered YOLOv4-tiny method. In order to tackle the challenges posed by existing methods concerning their high resource demand and poor real-time detection, we trained the YOLOv4-tiny model such that it achieved efficiency in terms of both high precision and faster inference speeds. We also notice that most abnormal movements during motion can be distinguished by a motion-triggering mechanism that isolates moving objects from static ones for analysis and detection. Experimental results indicate the potential benefits of the motion-triggering mechanism on moving object detection, and SpeedyGuard is 6.1 % faster and improved in real-time detection.
Authors: Yi Cao, Yunchong Long, Boyuan Yang, Peng Han
Institution: Wuhan University of Technology
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4409965513