README_notes.md

countYourBirds

Specify config.yaml

• Need to do in terminal if pip already installed: pip install pyyaml
• get an account on https://opensensemap.org/ and specify the user credentials in the config.yaml
• choose a name and coordinates for your CountYourBirds Box and specify them in the config.yaml

Image recognition live

First installation like described here: https://tutorials-raspberrypi.de/raspberry-pi-objekterkennung-mittels-tensorflow-und-kamera/

With this model: https://tfhub.dev/tensorflow/lite-model/ssd_mobilenet_v1/1/metadata/1?lite-format=tflite

(Other model for birds, squirrel and raccoons named in this tutorial: https://github.com/EdjeElectronics/TensorFlow-Lite-Object-Detection-on-Android-and-Raspberry-Pi)

Steps to activate, when you have already created the environment like in the description:

$ source tflite1-env/bin/activate
$ cd tflite1
$ python3 TFLite_detection_webcam.py --modeldir=Sample_TF_Lite_Model

Run image recognition file

1. Steps to activate, when you have already created the environment like in the description:

$ source tflite1-env/bin/activate
$ cd tflite1
$ python3 TFLite_detection_webcam.py --modeldir=Sample_TF_Lite_Model

2. Run folowing command add the name of your image at name.jpg

$ python TFLite_detection_image.py --modeldir=Sample_TF_Lite_Model --image name.jpg

Bird classification

Basis workflow found here: https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/examples/python/label_image.py By using this model: https://tfhub.dev/google/lite-model/aiy/vision/classifier/birds_V1/3

Motion detection

Use this description to start https://www.bouvet.no/bouvet-deler/utbrudd/building-a-motion-activated-security-camera-with-the-raspberry-pi-zero

Have a look at motion.conf settings: e.g. decrease event_gap to 1 or 2 Possible configuration settings can be found here: https://motion-project.github.io/4.3.1/motion_config.html

start motion with

• $ sudo motion -c /etc/motion/motion.conf or
• $ motion -c ~/.motion/motion.conf

stop with:

• $ sudo service motion stop

see http://raspberrypi:8080/ for motion and camera control

settings:

• trigger image recognition script with e.g. on_event_end ..\script.py

Start Detection

• python setup.py
• set directory in reset.py and opensensemapAPI.py
• motion motion.conf

Create own model

Tutorial: https://tensorflow-object-detection-api-tutorial.readthedocs.io/en/latest/training.html

Next steps

• get screenshots of several timesteps of the recorderd video to make them ready for classification
• video/ screenshot gets classifed by tensorflow (in the motion.conf you can define a script which should be excecuted if there is a picture or video is saved)
• send an email or mattermost message if there was a bird recognition (https://www.bouvet.no/bouvet-deler/utbrudd/building-a-motion-activated-security-camera-with-the-raspberry-pi-zero)
• classify the birds by species
• set up the raspberry pi at the balcony to produce first videos (using motion) + define settings for usable images
• think about infrared recognition because by camera it is very energy consuming (https://randomnerdtutorials.com/raspberry-pi-motion-detector-photo-capture/)

16.01

• send pictures with most confidence
• upload pcitures and count to sensebox

31.01

• write complete guideline
• update requirements.bash
• test on fresh py
• allow more sensors
• export more settings to yaml
• set description on creation
• make repository public
• upload model?
• catch errors