ros2_intel_movidius_ncs

This project is a ROS2 wrapper for the Movidius Neural Compute Stick (NCS) API, enabling real-time deep learning inference at the edge within the Robot Operating System 2 (ROS2) framework. It allows robots and edge devices to perform object classification and detection using pre-trained CNN models, leveraging the low-power Movidius Vision Processing Unit (VPU).

Key Features

• Image Classification Service — A ROS2 service for classifying objects in static image files using supported CNN models.
• Video Stream Publisher — A ROS2 publisher that performs object classification and detection on live video streams from RGB cameras.
• Multi-Framework Model Support — Supports CNN models from both Caffe and TensorFlow frameworks, including AlexNet, GoogLeNet, SqueezeNet, Inception variants, MobileNet, and MobileNetSSD.
• Demo Applications — Includes ready-to-run demos showcasing image and video classification/detection capabilities.

Philosophy

The project is designed to integrate deep learning acceleration hardware seamlessly into the ROS2 ecosystem, making edge AI accessible for robotics applications with efficient, low-power inference.