usage: tf_app.py [-h] [-b BLOB] [-i INPUT] [-r] [-s IMAGESET] [-f FREQ] [-t THRESHOLD]
optional arguments:
-h, --help show this help message and exit
-b BLOB, --blob BLOB blob type (s3 or azure). Defaults to s3
-i INPUT, --input INPUT
path to video feed. Format: "http://feedone,feed1_name http://feedtwo,feed2_name"
-r, --rt enable TensorRT
-s IMAGESET, --imageset IMAGESET
Imageset to use (coco or oid). Defaults to coco
-f FREQ, --freq FREQ Analysis frequency in seconds. Defaults to 10
-t THRESHOLD, --threshold THRESHOLD
detection threshold. Defaults to 0.40
- OpenCV captures a video feed (a webcam for example)
- Image is passed to Tensorflow for object detection analysis (by default this happens every 10 seconds)
- Tensorflow determines what objects are present in the image
- If objects are present, the image is uploaded to Amazon S3 (images retained for 35 days)
- Object labels and image URL are sent via MQTT to a broker, message is collected by Node-RED, which forwards the data to InfluxDB
- Grafana graphs the labels, attaching the image URL to each data point as metadata
- Object labels graphed in time series
- Image URL metadata attached to each data point
- Image with Tensorflow object detection overlay can be display via the graph link