YOLO_robocup

YOLOv3 Matlab Implementation for Robocup challenge.

Objectives

• Design a YOLOv2 Architecture for Robocup 2021 Manipulator challenge, that recognizes bottles and cans in a simulated environment.
• Train the neural network for a robust prediction.

Dependencies

It's required to have Image Processing Toolbox, Computer Vision Toolbox and Deep Learning Toolbox

Preprocess Data

Multiple images were taken in the Gazebo environment from the perspective. of the RGB-D Camera of the Genova3 Robot Arm. The input image is a 480x270x3 (RGB) resized to 270x270x3. A DataStore is created for the Labels and the Images , with the Image Labeler Matlab App

YOLOv3 Implementation

Based on a SqueezeNet Feature extraction.

## Authors CEIMTUN-RAS 2021 - Universidad Nacional de Colombia.