GRIP vision processing on an Intel Galileo

This repository contains documentation and source code for running GRIP vision processing pipelines on an Intel Galileo embedded computer. The primary audience is FIRST FRC teams interested in using the retro-reflective targets on the game field as a guide for autonomous robot operation.

This Wiki is divided into several guides:

• Why use the Intel Galileo for vision processing?
• Intel Galileo setup
• Compiling and installing GRIPonGalileo
• Creating the GRIP pipeline
• Microsoft Lifecam manual exposure details

As an introduction, here are the main components of the project:

• GRIPonGalileo is an program written in C++ that runs a vision processing pipeline developed with the GRIP graphical tool. It fetches frames from a USB webcam and converts them to the format needed by the vision pipeline. The project includes specialized webcam support for the Microsoft Lifecam HD-3000, often used by FRC teams. It is assumed that the last step of the pipeline is to find or filter contours corresponding to the vision targets. After invoking the pipeline, GRIPonGalileo sends a report of the contours to the RoboRIO via Network Tables.

  The last step mirrors the "NT Publish ContoursReport" building block that's included in GRIP. As a result, GRIPonGalileo is compatible with the pipelines you develop on a laptop or desktop computer. Typically, pipelines are written and tested using a computer and then exported to the Galileo to run in a self-contained manner on the robot. See Creating the GRIP pipeline for more details on the workflow.

• GRIP is a graphical application for designing and debugging vision processing pipelines built on OpenCV. Vision inputs can be configured for webcams, network cameras, or collections of images taken in the field. As filtering operations are added and adjusted, the results can be seen immediately. The output is generated source code that implements the pipeline in Java, C++, or Python. With GRIP, computer vision is much more approachable for novices and significantly more convenient for experienced OpenCV users. (https://github.com/WPIRoboticsProjects/GRIP/blob/master/README.md)

• Intel Galileo is a single-board computer built for embedded and IOT applications. It is used in this project to offload the CPU-intensive image processing work from the main control computer on the robot. This removes the risk of sluggish or jittery robot response resulting from vision pipelines that take too long to execute. It is inexpensive (~$50) and comes with exactly the interfaces and OpenCV libraries needed for this use case. (https://software.intel.com/en-us/iot/hardware/galileo)

• Microsoft Lifecam HD-3000 is a USB webcam recommended by FIRST and supported on the Mac, Windows, and Linux. A key element of this project is custom-written software for obtaining images from the Lifecam with proper exposure. (The automatic exposure levels for this camera are far too high for FRC vision processing but manual controls are available via the driver API.)