Skip to content

Latest commit

 

History

History
99 lines (86 loc) · 4.6 KB

object_zone_count.md

File metadata and controls

99 lines (86 loc) · 4.6 KB
Raw

Zone Event Detection

The Zone Event Detection AI Skill is used to determine if detected objects reside in polygons that this Skill takes as inputs.

Parameters

The extension takes the following parameters. All parameters are optional for the pipeline to run.

zones

A list of zone definitions which are objects containing the following fields.

  • name : the name of the zone for use in event reporting.
  • polygon : A list of four vertices (a tuple of x,y coordinates) which make up the bounds of the polygon.

If this parameter is not set, the extension defaults to an empty list and will not check for zone detections.

"zones": [
    {
        "name": "Zone1",
        "polygon": [[0.01,0.10],[0.005,0.53],[0.11,0.53],[0.095,0.10]]
    },
    {
        "name": "Zone2",
        "polygon": [[0.14,0.20],[0.18,0.67],[0.35,0.67],[0.26,0.20]]
    },
    {
        "name": "Zone3",
        "polygon": [[0.40,0.30],[0.50,0.83],[0.85,0.83],[0.57,0.30]]
    }
]

enable_watermark

A boolean flag that defaults to false. If set to true marks the vertices of all zones and annotates the first vertex with the zone name.

log_level

The logging level defined as a string. Defaults to "INFO".

Event Output

If a tracked object crosses any of the lines, an event of type object-zone-count will be created with the following fields.

  • zone-name: name of the associated line
  • related-detections: array containing the array indices of the detected objects in the zone
  • status : array containing the status of the detected objects in the zone. Values can be "within" to indicate that the object is fully within the zone or "intersects" if it touches any of the vertices.
  • zone-count : total number of objects in the zone
{
   "event-type":"zoneCrossing",
   "zone-name":"Zone1",
   "related-objects":[
      0, 1, 4
   ],
   "status":[
      "within",
      "within",
      "intersects"
   ],
   "zone-count" : 3
}

Example Run

Intel(R) Deep Learning Streamer (Intel(R) DL Streamer) Pipeline Server comes with an example configuration for object-zone-count

  1. Build & Run the Pipeline Server

  2. Run object-zone-count pipeline with vaclient using example parameter file:

    vaclient/vaclient.sh run object_detection/object_zone_count https://github.com/intel-iot-devkit/sample-videos/blob/master/people-detection.mp4?raw=true --parameter-file vaclient/parameter_files/object-zone-count.json

    You will see events among the detections in vaclient output:

    Timestamp 45000000000
- person (0.76) [0.28, 0.15, 0.42, 0.72]
Event: event-type: object-zone-count, zone-name: Zone2, related-objects: [0], status: ['intersects'], zone-count: 1
Event: event-type: object-zone-count, zone-name: Zone3, related-objects: [0], status: ['intersects'], zone-count: 1
Timestamp 45166666666
- person (0.57) [0.24, 0.16, 0.35, 0.69]
Event: event-type: object-zone-count, zone-name: Zone2, related-objects: [0], status: ['intersects'], zone-count: 1

Watermark Example

  1. Open the example configuration and add enable_watermark as follows:

    "object-zone-count-config": {
    "zones": [
        <snip>
    ],
    "enable_watermark": true
}

  2. Rebuild and Run the Pipeline Server with additional flag --enable-rtsp

    ./docker/run.sh -v /tmp:/tmp --enable-rtsp

  3. Run object-zone-count pipeline with vaclient using example parameter file with additional parameter rtsp-path. Note that rtsp-path is set to pipeline-server, this path is what will be used to view the rtsp stream:

    vaclient/vaclient.sh run object_detection/object_zone_count https://github.com/intel-iot-devkit/sample-videos/blob/master/people-detection.mp4?raw=true --parameter-file vaclient/parameter_files/object-zone-count.json --rtsp-path pipeline-server

  4. Open up a media player with network stream viewing (VLC for example) and connect to rtsp:://<host ip>:8554/pipeline-server. The stream is real time so you might want to setup your media player ahead of time. You will see people-detection.mp4 with an overlay of points. Each zone has a start point which has a label of the zone name. Other points of the zone are not labeled. If an object intersects or is within a zone the label is updated to reflect that.

    object_zone_count_watermark