Flask Video Streaming for Object Detection

The video streaming over flask is referred from https://github.com/miguelgrinberg/flask-video-streaming, the Montion JPG idea. More details about it please refer to video streaming with Flask and follow-up of the previous article Flask Video Streaming Revisited.

The second part is object detection on tensorflow. We combine them in order to deploy on the device or the environment without display, for example, IoT devices or container, etc.

The model is ssd_mobilenet_v1_coco referred from https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md.

Quick Start

Developing Mode

• Run a simple streaming example from a video file over flask.

$ python app.py

• Run a streaming example from a web camera over flask. In windows, run set CAMERA=opencv first.

$ CAMERA=opencv python app.py

• Run a realtime object detection example from a web camera over flask. In windows, run set CAMERA=obdet first.

$ CAMERA=obdet python app.py

After start the service, you can surf the web on http://localhost:5000.

Production Mode

Here we use Gunicorn tool to deploy our services, notice that windows platform did not support it.

Notice if you run the following scripts using a external webcam on macbook, it might go wrong due to AVCaptureSession warning: Session received the following error while decompressing video: Error Domain=NSOSStatusErrorDomain Code=-12903.

• [ubuntu] Run based on the threading worker.

# --threads: allow how many clients access to the flask server
# --worker-class: default is sync
$ CAMERA=opencv gunicorn --threads 5 --workers 1 --bind 0.0.0.0:5000 app:app

• [ubuntu] Run based on un-sync concurrency.

# worker type: eventlet, gevent, tarnado, etc.
$ CAMERA=opencv gunicorn --worker-class gthread --threads 5 --workers 1 --bind 0.0.0.0:5000 app:app
$ CAMERA=obdet gunicorn --worker-class gthread --threads 5 --workers 1 --bind 0.0.0.0:5000 app:app

• [Macbook] Run the script using internel camera only, not for externel camera.

$ CAMERA=opencv gunicorn --worker-class gthread --threads 5 --workers 1 --bind 0.0.0.0:5000 app:app

After start the service, you can surf the web on http://<IP-address>:5000.