Sensor and Neural Data Platform (SAND)

This project aims to make working with camera-streams (or video-streams in general) easy, by providing an extensible framework to read, manipulate and republish images.

It was originally designed to manage an Object-Detection-System we setup on a tri-modal crane. The focus of this was to detect humans (and biomass in general) to make the processes at container-harbors safer in general.

While developing we already recognized, that it could also be useful for generic usages, especially just recording some simple streams or do some light work, like republishing streams.

How to setup

In general the installation process is quite simple. We described the simple terms here, but if you need a reference of some sort, we can recommend taking a look at our Dockerfile that we use for CI-purposes.

Prerequisites

System Dependencies

You need those in any case:

• gcc
• build-essential
• libcap-dev
• python3-dev

Neural Network

If you want to use a demo yolox please download the weights for YOLOX-X (trained on the COCO dataset) from here and either put them directly in ./config/detector_models/yolox/yolox_x_coco or use a symlink.

Development setup

If you only want to use SAND directly, you can skip this section. Although we can very much recommend poetry in general for your own python projects.

We use poetry to manage our dependencies. To install poetry you can use (more infos here):

$ curl -sSL https://install.python-poetry.org | python -

Our dependencies are documented in the pyproject.toml, the explicit versions with hashes for the libraries you can find in poetry.lock.

In the following section you will see what different variants of the system we actually have.

Installation

Attention: As "just" a user you probably don't want to install poetry or any dev-dependencies. Therefore you should always look at the pip install variant.

We have a couple of variants for the system, depending on what you want to do with it. They can also be combined with each other if you want to use multiple features. For development and especially unittesting you will need all extras:

• neural

  $ poetry install --extras="neural"
  $ pip install python-sand[neural]
  

  This installs all the dependencies that have something to do with machine learning or neural networks. The main dependency here is the MLCVZoo which provides us with an easy interface to work with inference and the results of it in general. If you want to configure some kind of object-detection this and the corresponding component in our system (NeuralNetwork) are a good starting point to look into.

• metric

  $ poetry install --extras="metric"
  $ pip install python-sand[metric]
  

  This concerns everything around an influx-db and metrics with a grafana-dashboard. So if you want to monitor performance of basically everything in our system you can start here. This absolutely needs a running MQTT-Broker that the system can connect to. For development and "light" running we have a mosquitto docker container setup in the docker-compose.yml. docker-compose is also installed via our dev-dependencies which get installed regardless of extras if you install it via poetry. You will also need an influx-dbv2.

  $ docker-compose up -d mqtt influxdb grafana
  

• publisher

  $ poetry install --extras="publisher"
  $ pip install python-sand[publisher]
  

  This basically gives you the tools to setup a basic flask server to deliver in our case a static website where you can watch your streams in a basic dashboard.

System

If you have installed python-sand in your environment it provides a couple of binaries, that you can use to start different parts of the system.

• sand

  This is the main starter for the "normal" system. It has a couple of options so be sure to take a look (via the normal sand --help)

Long-term System

If you want to run install it on the actual system where it should run long-term, we opted for a systemd-service to make starting/stopping very easy and also the logging gets easier. You probably still want to adapt it slightly to use your specific config or use additional links to match the default config name. You find the systemd file in this repository, it will not come bundled in the python artifact.

Installation:

# cd /etc/systemd/system
# ln -s /path/to/sand/sand.service .

After that you can start/stop it via:

# systemctl start sand

Also the logs on the INFO-Level are routed through the journal, which is why you can also read most of the logs via:

# journalctl -u sand

FAQ

More like "We asked them ourselves at one point, and tried to find a spot where to save the knowledge".

How can you reset the admin password in grafana docker container?

$ docker exec CONTAINER_ID grafana-cli admin reset-admin-password admin