mooseetws / mooseetws-tensorflow-detection

Tensorflow object detection and risk reporting running on RaspberryPI

mooseetws-edge-computing

Object detection based on tensorflow serving REST API, risk identification and reporting.

Containers

• Tensorflow serving

  • Serve pre-trained models using tensorflow serving REST API
  • Used ssdlite_mobilenet_v2_coco_2018_05_09
  • API endpoint:

  curl --request GET \
    --url http://localhost:8501/v1/models/ssdlite_mobilenet_v2_coco_2018_05_09
  

  • Predict API:

  curl --request POST \
    --url http://localhost:8501/v1/models/ssdlite_mobilenet_v2_coco_2018_05_09:predict \
    --header 'content-type: application/json' \
    --data '{  "instances": [ "foo", "bar", "baz" ]}'
    
  

• Camera feeding

  • Periodically scan usb camera and use tensorflow API to detect potential threats.
  • Post notification to cloud service when risks are identified.

Reference:

• Tensorflow serving for ARM: https://hub.docker.com/r/emacski/tensorflow-serving
• Tensorflow serving documentation: https://www.tensorflow.org/tfx/serving/docker