People detection on 2D laser range finder data using deep learning and machine learning

Abstract

This work presents a machine learning based study on people detection using 2D Laser Range Finders (LRFs) combined with deep learning methodologies, aimed at enhancing mobile robot capabilities in various environmental conditions. The study introduces a novel integration of a monocular camera with an LRF on a mobile robot to improve the accuracy and efficiency of detecting and tracking people. By employing deep learning models such as CenterNet, the system leverages both image and 2D range data to facilitate automatic labeling of datasets, crucial for training robust classification algorithms. In order to achieve the best classifier, two experimental studies are introduced in this work. The former is carried out in a simulated environment and the latter in real-world, office-like environments. In simulations, various machine learning models are trained and evaluated, showing significant results in distinguishing human legs from other objects. The transition to real-world testing underscores the challenges and adaptations necessary to achieve high accuracy and reliability in dynamic settings. The XGBoost model emerged as the most effective classifier in our study, achieving the highest scores in accuracy, precision, recall, and F1-score, outperforming other methods across these key metrics. This work aims to advance the field of 2D LRF based people detection and also proposes a solution for real-time applications, balancing precision and computational efficiency. Experimental results from both simulated and real-world environments demonstrate the system's effectiveness.