Position and velocity determination of people aimed at study of evacuations

This study describes the application of algorithms on video sequences with the purpose of detecting and identifying individuals moving in crowds (even with occasional occlusions), calculating their position, speed, and trajectory. To achieve this, the YOLO neural network was utilized for people detection, and the ORB descriptor was used for their tracking. Additionally, a Kalman filter was employed to estimate the position and velocity of each detected person. This configuration was applied to various videos, achieving excellent results in terms of the number of people detected and the accuracy in determining the variables. The obtained outcomes are comparable in precision and processing times to those sought in the development of an automated control system applied to people circulation.