Skip to content
A Raspberry Pi powered, distributed (edge) computing camera setups that runs a Tensorflow object detection model to determine whether a person is on the camera. A plugin model allows actions based on the detection, such as playing audio, turning on lights, or triggering an Arduino.
Python Jupyter Notebook Shell
Branch: master
Clone or download
christian
christian Merge branch 'development' into 'master'
Development

See merge request christian/scarecrow!4
Latest commit c89cc23 Dec 24, 2019

README.md

Scarecrow-Cam

A Raspberry Pi powered, distributed (edge) computing camera setups that runs a Tensorflow object detection model to determine whether a person is on the camera. The Raspberry Pi is used for video streaming and triggering actions (such as playing audio, turning on lights, or triggering an Arduino), whereas a server or laptop runs the object detection. With a suitable TFLite installation, this can happen locally on the Raspberry as well.

Based on the detection criteria, a plugin model allows to trigger downstream actions.

Based on my blog.

Architecture

Architecture

Sample

Side note: The setup shown here only fits the use-case of edge to a degree, as we run local detection on a separate machine; technically, the Raspberry Pi is capable of running Tensorflow on board, e.g. through TFLite or esp32cam.

You can change this behavior by relying on a local tensorflor instance and having the ZMQ communication run over localhost.

Requirements

This project requires:

  • A Raspberry Pi + the camera module v2 (the client)
  • Any Linux machine on the same network (the server)

Pi

Install

A helper script is available:

Use a virtual environment

python3 -m venv env
source env/bin/activate
bash ./sbin/install_tf_vidgear.sh [server/client]

Please see INSTALL.md for details.

Configuration and data

Edit the conf/config.ini with the settings for your Raspberry and server.

For playing audio, please adjust conf/plugins.d/audio.ini.

[Audio]
Path=../audio_files

For an appropriate path.

If you want to change the model, please check the Model Zoo. The blog article used the outdated ssd_mobilenet_v1_coco_2017_11_17.

[Tensorflow]
ModelUrl=ssd_mobilenet_v3_large_coco_2019_08_14

Run

On the raspberry

python3 -m client.sender --input 0 # for picam
python3 -m client.sender --input '/path/to/video' # for local video

On the server

python3 -m server.receiver

Plugins

A plugin model allows to trigger downstream actions. These actions are triggered based on the configuration.

Plugins can be enabled by setting the following in config.ini:

[Plugins]
Enabled=audio
Disabled=

Currently, the following plugins are avaibale:

Plugin Description Requirements Configuration Base
audio Plays audio files once a person is detected Either playsound, pygame, or omxplayer conf/plugins.d/audio.ini ZMQ
store_video Stores video files on the server, with a defined buffer or length Path and Encoding settings conf/plugins.d/store_video.ini ZServerMQ

How to contribute

This project is in an early state of development. Therefore, there are several open items that need to be covered. Please see TODO for details.

License

This project is licensed under the GNU GPLv3 License - see the LICENSE file for details.

You can’t perform that action at this time.