Moment

Autonomous Camera Tracking - https://medium.com/mmnt

1. Installation

1.1 Jetson Setup

Ensure Jetson has Jetpack installed for Cuda (using Jetpack 3.3).

For verification, the following should output Cuda release 9.0.
```
/usr/local/cuda/bin/nvcc --version
```

Set up and activate virtual environment for Python 3.

sudo apt update
sudo apt install python3-dev python3-pip
sudo pip3 install -U virtualenv  # system-wide install

mkdir ~/venvs
virtualenv --system-site-packages -p python3 ~/venvs/mmnt
source ~/venvs/mmnt/bin/activate # to activate the virtual environment
deactivate # to deactivate the virtual environment

Install Tensorflow from a pre-built wheel (using TF 1.11.0 TRT Pyton 3.5).

# After downloading the .whl file
pip install tensorflow-1.11.0-cp35-cp35m-linux_aarch64.whl

Install OpenCV (using OpenCV 3.4.5).
Clone tf-openpose

Follow the install instructions on the readme. The package install flow seemed to work better than the regular install flow.
Clone the MMNT repo
Setup for mmnt environment

sudo apt-get update sudo pip install pyusb sudo pip install pyserial brew install libusb ```

1.2 Mic Setup

Install dependencies with the following command:

sudo apt-get update
sudo pip install pyusb
sudo pip install pyserial
brew install libusb

Issues

Running pyserial (for the arduino) on a VM does not have required permissions

After enabling the usb device in the VirtualBox VM settings, the serial output corresponding to the arduino should be /dev/ttyACM0.

Run the following to gain permission: sudo chmod 666 /dev/ttyACM0

Running pyusb for the mic on a VM does not have the required permissions

Find the usb address of the mic with the following: lsusb

Then, use the address to gain permission: sudo chmod o+rw /dev/bus/usb/001/00<USB ID>

2. Execution

To run the system, simply execute the python main:

cd ../mmnt/src/
python main.py

NavidMia/mmnt