Solução de aprendizado de máquina para contagem de pessoas em ambientes internos

Abstract

This work presents the development of a computer vision system based on machine learning for automated people counting in indoor environments. The main motivation for this proposal is its application in automation solutions aimed at optimizing energy consumption by intelligently controlling devices such as air conditioning and lighting. The system was developed using Google Colab, Keras, and the YOLO model from Ultralytics. The results demonstrate the effectiveness of the proposed approach, with the final model achieving a mean absolute error (MAE) of 0.099, even in challenging scenarios with high individual density and overlapping. This performance highlights the model’s ability to handle complex visual contexts, making it a promising alternative for practical applications in automation and indoor monitoring.