Video Analytics Serving provides the structure and key components needed to bootstrap development of visual processing solutions. Video Analytics Serving has been designed to simplify the deployment and use of hardware optimized video analytics pipelines. A reference service implementation exposes RESTful interfaces that make use of example pipelines to get you started.
These endpoints and pipelines may be customized to negotiate the inputs and outputs appropriate to many use cases. Developers may choose to customize and then execute pre-defined video analytics (VA) pipelines in either GStreamer or FFmpeg.
Each VA pipeline type defines the semantics of its customizable parameters. These parameters are included in requests to start a pipeline and will influence the runtime behavior of a VA pipeline. In this way VA pipeline developers may define named and versioned VA pipelines and expose them to users via simple RESTful interfaces.
IMPORTANT: Video Analytics Serving is provided as a pre-production sample.
The project provides a reference architecture with straightforward examples to accelerate your implementation of a solution. However, it is not intended for production without modification. In addition to modifying pipelines and models to fit your use cases, you must harden security of endpoints and other critical tasks to secure your solution.
Path | Description |
---|---|
GET /models |
Return supported models. |
GET /pipelines |
Return supported pipelines |
GET /pipelines/{name}/{version} |
Return pipeline description. |
POST /pipelines/{name}/{version} |
Start new pipeline instance. |
GET /pipelines/{name}/{version}/{instance_id} |
Return pipeline instance summary. |
GET /pipelines/{name}/{version}/{instance_id}/status |
Return pipeline instance status. |
Video Analytics Serving includes two sample analytics pipelines for GStreamer and FFmpeg. The GStreamer sample pipelines use plugins for CNN model-based video analytics utilizing Intel OpenVino. When building with Docker, these plugins will be built and installed inside the Docker image. You can find documentation on the properties of these elements here.
Pipeline | Description | Example Request | Example Detection |
---|---|---|---|
/pipelines/object_detection/1 | Object Detection | curl localhost:8080/pipelines/object_detection/1 -X POST -H 'Content-Type: application/json' -d '{ "source": { "uri": "https://github.com/intel-iot-devkit/sample-videos/blob/master/bottle-detection.mp4?raw=true", "type": "uri" }, "destination": { "type": "file", "uri": "file:///tmp/results.txt"}}' | {"objects": [{"detection": {"bounding_box": {"x_max": 0.8820319175720215, "x_min": 0.7787219285964966, "y_max": 0.8876367211341858, "y_min": 0.3044118285179138}, "confidence": 0.6628172397613525, "label": "bottle", "label_id": 5}}], "resolution": {"height": 360, "width": 640}, "source": "https://github.com/intel-iot-devkit/sample-videos/blob/master/bottle-detection.mp4?raw=true", "timestamp": 7407821229} |
/pipelines/emotion_recognition/1 | Emotion Recognition | curl localhost:8080/pipelines/emotion_recognition/1 -X POST -H 'Content-Type: application/json' -d '{ "source": { "uri": "https://github.com/intel-iot-devkit/sample-videos/blob/master/head-pose-face-detection-male.mp4?raw=true", "type": "uri" }, "destination": { "type": "file", "uri": "file:///tmp/results1.txt"}}' | {"objects": [{"detection": {"bounding_box": {"x_max": 0.5859826803207397, "x_min": 0.43868017196655273, "y_max": 0.5278626084327698, "y_min": 0.15201044082641602}, "confidence": 0.9999998807907104, "label": "face", "label_id": 1}, "emotion": {"label": "neutral", "model": {"name": "0003_EmoNet_ResNet10"}}}], "resolution": {"height": 432, "width": 768}, "source": "https://github.com/intel-iot-devkit/sample-videos/blob/master/head-pose-face-detection-male.mp4?raw=true", "timestamp": 133083333333} |
(1) Install docker engine.
(2) Install docker compose, if you plan to deploy through docker compose. Version 1.20+ is required.
Video Analytics Serving may be modified to co-exist in a container alongside other applications or can be built and run as a standalone service.
To get started, build the service as a standalone component execute the following command
./build.sh
After a successful build, run the service using the included script
./run.sh
This script issues a standard docker run command to launch the container. Volume mounting is used to publish the sample results to your host.
With the service running, initiate a request to start a pipeline using the following commands.
(1) curl localhost:8080/pipelines/object_detection/1 -X POST -H 'Content-Type: application/json' -d '{ "source": { "uri": "https://github.com/intel-iot-devkit/sample-videos/blob/master/bottle-detection.mp4?raw=true", "type": "uri" }, "destination": { "type": "file", "uri": "file:///tmp/results.txt"}}'
(2) tail -f /tmp/results.txt
{"objects": [{"detection": {"bounding_box": {"x_max": 0.9027906656265259, "x_min": 0.792841911315918, "y_max": 0.8914870023727417, "y_min": 0.3036404848098755}, "confidence": 0.6788424253463745, "label": "bottle", "label_id": 5}}], "resolution": {"height": 360, "width": 640}, "source": "https://github.com/intel-iot-devkit/sample-videos/blob/master/bottle-detection.mp4?raw=true", "timestamp": 39854748603}