Computer Vision with Tensorflow - NIURoverTeam/RoverCoreOS GitHub Wiki

Dependencies

Before you install Tensorflow and the ROS package in the next step, make sure you have ROS and the required Realsense software installed. The home page here has the instructions for that.

Installation

The two main things you're going to need to install are TensorFlow and the ROS package that will work with it.

Instructions on how to install TensorFlow can be found here. Double check to make sure you're following the instructions for Python 2.7, not 3.

The ROS Object Detection package can be found here.

Setting Parameters

You're going to need to pick a model for the package to use. Currently, we're using a model named faster_rcnn_inception_v2_coco_2018_01_28. Different models can be installed from the Model Zoo.

Model files should be extracted into catkin_ws/src/cob_people_object_detection_tensorflow/src/object_detection

To update the model:

cd into catkin_ws/src/cob_people_object_detection_tensorflow/launch and open launch/cob_people_object_detection_tensoflow_params.yaml in a file editor.

From there, just copy and paste the new model name into the file.

Note: If it's your first time running this package, you'll probably need to update the topics for your camera. For the RealSense, the RGB image topic will likely need to be changed to /camera/color/image_raw.

Running

First, turn the RealSense on with roslaunch realsense2_camera rs_camera.launch

Next, in a new terminal, start the TensorFlow virtual environment with source ./venv/bin/activate.

After that, you're good to run the Object Detection package with roslaunch cob_people_object_detection_tensorflow alltogether.launch.