YOLO Object Detectors for Robotics -- a Comparative Study

TL;DR

This paper compares different YOLO object detectors to evaluate their effectiveness and robustness for robotic vision tasks using custom and COCO datasets with distortions.

Contribution

It provides a comprehensive validation of YOLO models' applicability to robotic object detection, including robustness analysis under various distortions.

Findings

Certain YOLO versions perform better in robotic environments.

Distortions significantly affect detection accuracy.

Training/testing configurations influence model robustness.

Abstract

YOLO object detectors recently became a key component of vision systems in many domains. The family of available YOLO models consists of multiple versions, each in various variants. The research reported in this paper aims to validate the applicability of members of this family to detect objects located within the robot workspace. In our experiments, we used our custom dataset and the COCO2017 dataset. To test the robustness of investigated detectors, the images of these datasets were subject to distortions. The results of our experiments, including variations of training/testing configurations and models, may support the choice of the appropriate YOLO version for robotic vision tasks.