A Real-time, scalable and extensible object filtering and detection system using kinect sensor and ROS2 foxy

Abstract

Object detection and filtering based on shape and color is an important capability for many robotics applications. For example, sorting objects by shape and color is a common industrial application. Service robots also need to detect and track objects based on visual properties. While powerful deep learning approaches like YOLO have emerged for general object detection, they require large datasets and extensive training. A simple shape and color filtering provide a lightweight and customizable alternative. This work aims to provide a real-time modular and lightweight system that can identify objects of basic shapes and colors and allow extensibility of functionality by incorporating more custom color and shape filters. The proposed system for real-time shape and color filtering using a Microsoft Kinect RGB-D sensor and Robot Operating System (ROS2) can identify an array of regular shapes like circles, rectangles, and triangles over a spectrum of different colors. New shape and color filters can be added dynamically at runtime thanks to the modular, ROS-2-based implementation.