DEVELOPMENT OF A COMPUTER VISION SYSTEM USING YOLOV8 FOR DETECTING AND COUNTING THE NUMBER OF PEOPLE ENTERING AND EXITING.

Abstract

This study employed YOLOv8, an advanced neural network, to develop a real-time artificial vision system for detecting and counting people. A total of 7250 images were collected using Roboflow to train the model, enhancing its accuracy through data augmentation techniques. The training process leveraged a Tesla T4 GPU on Google Colab for accelerated processing. The system achieved an average accuracy of 94.6%, with peaks reaching 100% at specific times, albeit encountering some false positives. These findings underscore YOLOv8's effectiveness in enhancing security and crowd management, suggesting future enhancements in model confidence and image quality could further improve performance.