Real-time gender and people tracking using YOLO
Ziyaad Muhammad Peerun, Raj Kishen Moloo · 2024
The objective of this research was to use the YOLO API and other vision-based tools to detect gender and keep track of the number of people in real-time. The CNN Darknet-53 was used for the gender classification. Two models namely YOLOv3 and YOLOv3-TINY were tested for gender detection to show the trade-off between speed and accuracy. An additional YOLOv3 model was also tested for people detection and tracking. The Google Colaboratory was used to train the network using a high-end GPU Nvidia Tesla K80 12GB on images from the ‘OpenImagesv5’ dataset. The YOLOv3 and YOLOv3-TINY models achieved commendable accuracy on the training set and people detection. The models were deployed using the OpenCV DNN module as well as the Flask web application development framework. The web application allows end-users to view the results in real-time on a web browser. Ultimately, the system was tested on the campus of the University of Mauritius and achieved an average detection accuracy of $94 \%$