COMPARISON OF CHILD DETECTION SYSTEM WITH ARTIFICIAL INTELLIGENCE USING MOBILENET, YOLOV2 AND YOLOV3 FOR OBJECT DETECTION

DON PEREZ, LIAP (2019) COMPARISON OF CHILD DETECTION SYSTEM WITH ARTIFICIAL INTELLIGENCE USING MOBILENET, YOLOV2 AND YOLOV3 FOR OBJECT DETECTION. [Final Year Project Report] (Unpublished)

	PDF COMPARISON OF CHILD DETECTION SYSTEM WITH ARTIFICIAL24pgs.pdf Download (1MB)
	PDF (Please get the password by email to repository@unimas.my , or call ext: 3914 / 3942 / 3933) COMPARISON OF CHILD DETECTION SYSTEM WITH ARTIFICIALft.pdf Restricted to Registered users only Download (6MB)

Abstract

This research focuses on improving the child detection system by utilizing AI with recent versions of pre-trained models as an alternative of using sensors existing in the child detection system. Currently the problems experienced is the sensors used in the market, to prevent child heatstroke in automobiles, cannot accurately determines the occupant in position and whether the person is an adult or child. Comparing pre-trained models of object detection with AI such as Single-Shot Detector MobileNet (SSD MobileNet), YOLO version 2 (YOLOv2), and YOLO version 3 (YOLOv3) could suggests a more accurate and precise child detection system. The system with three different object detection models were tested experimentally to evaluate the speed, accuracy and precision. At the end of experiments, it is founded that YOLO able to detect custom objects the fastest which is less than a second. Also, YOLOv3 Tiny GPU has the best average score of detection which is 100% at the first 80 cm while SSD MobileNet has its highest average score for detection at 70 cm. At the best distance, which is 70cm, SSD MobileNet shows an acceptable result since there is no false detection, while YOLO shows perfect reproducibility result at 70 cm. In conclusion, YOLOv3 is the most suitable model to improve the framework of the child detection system.

Actions (For repository members only: login required)

View Item