Desarrollo de un modelo de aprendizaje automático para el seguimiento de objetos

Abstract

This graduation project presents a machine learning model that performs multiple object tracking (MOT) using visual-based tracking-by-detection approach. It uses state of the art realtime algorithms: YOLO (You Only Look Once) for detection and DeepSORT (Deep Simple Online and Realtime Tracking) for tracking. The model has been developed in Python in Google Colab environment and evaluated using KITTI (Karlsruhe Institute of Technology and Toyota Technological Institute) 2D MOT benchmark. The project contextualizes development of the model in the field of autonomous vehicles.