AWS DeepRacer object detection package

Overview

The object detection ROS package creates the object_detection_node, which is responsible for collecting sensor data (camera images) from sensor_fusion_pkg and running them through the object-detection model to find a specified object and provide a normalized delta of the found object from the target position. This delta value is published using a ROS publisher as DetectionDeltaMsg data. For more information, see the AWS DeepRacer Follow the Leader(FTL) sample project.

License

The source code is released under Apache 2.0.

Installation

Follow these steps to install the AWS DeepRacer object detection package.

Prerequisites

The AWS DeepRacer device comes with all the prerequisite packages and libraries installed to run the FTL sample project. For more information about the preinstalled set of packages and libraries on the AWS DeepRacer device, and about installing the required build systems, see Getting started with AWS DeepRacer OpenSource.

The object_detection_pkg specifically depends on the following ROS 2 packages as build and run dependencies.

1. deepracer_interfaces_pkg: This package contains the custom message and service type definitions used across the AWS DeepRacer core application, modified to support FTL sample project.

The following are the additional software and hardware requirements to get the object_detection_node to work on the AWS DeepRacer device.

1. Download and optimize the object-detection model: Follow the instructions to download and optimize the object-detection model and copy it to the required location on the AWS DeepRacer device.

2. Set up the Intel Neural Compute Stick 2 (optional): The object_detection_node provides functionality to offload the inference to a Intel Neural Compute Stick 2 connected to the AWS DeepRacer device. This is an optional setting that enhances the inference performance of the object-detection model. For more details about running inference on the Movidius NCS (Neural Compute Stick) with OpenVINO™ toolkit, see this video.

Attach the Neural Compute Stick 2 firmly in the back slot of the AWS DeepRacer, open a terminal, and run the following commands as the root user to install the dependencies of the Intel Neural Compute Stick 2 on the AWS DeepRacer device.

1. Switch to the root user:

        sudo su
   

2. Navigate to the OpenVino installation directory:

        cd /opt/intel/openvino_2021/install_dependencies
   

3. Set the environment variables required to run the Intel OpenVino scripts:

        source /opt/intel/openvino_2021/bin/setupvars.sh
   

4. Run the dependency installation script for the Intel Neural Compute Stick:

        ./install_NCS_udev_rules.sh
   

Downloading and building

Open a terminal on the AWS DeepRacer device and run the following commands as the root user.

1. Switch to the root user before you source the ROS 2 installation:

    sudo su
   

2. Source the ROS 2 Foxy setup bash script:

    source /opt/ros/foxy/setup.bash 
   

3. Set the environment variables required to run Intel OpenVino scripts:

    source /opt/intel/openvino_2021/bin/setupvars.sh
   

4. Create a workspace directory for the package:

    mkdir -p ~/deepracer_ws
    cd ~/deepracer_ws
   

5. Clone the entire FTL sample project on the DeepRacer device:

    git clone https://github.com/aws-deepracer/aws-deepracer-follow-the-leader-sample-project.git
    cd ~/deepracer_ws/aws-deepracer-follow-the-leader-sample-project/deepracer_follow_the_leader_ws/
   

6. Fetch the unreleased dependencies:

    cd ~/deepracer_ws/aws-deepracer-follow-the-leader-sample-project/deepracer_follow_the_leader_ws/
    rosws update
   

7. Resolve the dependencies:

    cd ~/deepracer_ws/aws-deepracer-follow-the-leader-sample-project/deepracer_follow_the_leader_ws/ && apt-get update
    rosdep install -i --from-path . --rosdistro foxy -y
   

8. Build the object_detection_pkg and deepracer_interfaces_pkg:

    cd ~/deepracer_ws/aws-deepracer-follow-the-leader-sample-project/deepracer_follow_the_leader_ws/ && colcon build --packages-select object_detection_pkg deepracer_interfaces_pkg
   

Using the object_detection_node

Although the object_detection_node is built to work with the FTL sample project, you can run it independently for development, testing, and debugging purposes.

Running the node

Configure the launch file to specify which device to use for inference (for more details, see the extended configuration section below). To launch the built object_detection_node as the root user on the AWS DeepRacer device, open another terminal on the device and run the following commands as the root user.

1. Switch to the root user before you source the ROS 2 installation:

    sudo su
   

2. Navigate to the FTL workspace:

File truncated at 100 lines see the full file