cam-event-detector

README

cam-event-detector

Camera image event detector for objects entering, moving, stopping or exiting the view.

Specialized on still security camera scenes.

What it does

In general, the event detector performs the following:

1. Reads continuously a camera image

2. Detects and tracks an object that is moving in the scene

3. Generates events related to the lifecycle of the tracked object: ENTERED, MOVED, STOPPED, EXITED

4. Schedules the event to be sent to a HTTP POST endpoint

5. Posts all the queued events json as soon as possible to the target HTTP endpoint. May have delays in case of network outages or insufficient bandwidth.

Usage

• create docker-compose.yml:

version: '3.7'

services:

  cam-event-detector:
    build: .
    image: flaviostutz/cam-event-detector
    ports:
      - 3000:3000
    restart: always
    environment:
      - CAM_ID=cam1
      - VIDEO_SOURCE_URL=http://195.252.80.186:9000/mjpg/video.mjpg
      - EVENT_POST_TARGET=http://localhost:8080/cam1/event

ENVs

• CAM_ID - 'cam_id' for the generated events

• VIDEO_SOURCE_URL - any video stream URL supported by OpenCV. ex.: http://cam1/v.mpeg; rtsp://cam2:554/stream

• EVENT_POST_TARGET - an HTTP POST endpoint whose detected events json contents will be sent asynchronously.

  • example: POST http://anotherserver/cam1/event

{
    "uuid": "ABDC-2342-BCCA",
    "cam_id": "cam1", //as in CAM_ID ENV
    "time": "2020-02-15T15:34:32",
    "type": "MOVED", //ENTERED, STOPPED, EXITED
    "bbox":{"x1":12,"y1":65,"x2":3,"y2":91},
    "speed": {"x":-5,"y":3},
    "image": "ABC3378BAAASA1", //JPEG image encoded in BASE64. if enabled in ENV EVENT_OBJECT_IMAGE_ENABLE
    "tracking": {
        "uuid": "ABDC-2342-BCCA",
        "keypoints": [
            {"ts":"2020-02-15T15:34:01", "bbox":{"x1":12,"y1":45,"x2":23,"y2":81}, "speed":{"x":-2,"y":1}},
            {"ts":"2020-02-15T15:34:01", "bbox":{"x1":12,"y1":65,"x2":3,"y2":91}, "speed":{"x":-5,"y":3}}
        ]
    },
    "scene": {
        "width": 800,
        "height": 600,
        "image": "AAC73621AAACCCDDD" //if enabled in ENV EVENT_SCENE_IMAGE_ENABLE
    },
}

• EVENT_OBJECT_IMAGE_ENABLE - if 'true', the cropped detected object image will be included in event payload. defaults to 'true'

• EVENT_SCENE_IMAGE_ENABLE - if 'true', the full scene image will be included in event payload. defaults to 'false'

• EVENT_MAX_KEYPOINTS - the max number of keypoints that will be included in event payload. defaults to '-1' (no limit)

Use OpenCV native windows for debugging

• Run the container with default CMD (/startup.sh) - comment your own

• Connect to the container with X export enabled: ssh -Y -p 2222 root@localhost

• cd /workspace - this is mounted to your desktop workspace, so all modifications in your editor will be present here

  • Map your local development workspace directory (where you git clone the projects you are working in) to /workspace

• go build -o /usr/bin/cam-event-detector && cam-event-detector --cam-id=test1 --video-source-url=rtsp://dummy-rtsp-relay:8554/stream --event-post-endpoint=http://dummy-event-receiver:3000/events

• go build -o /usr/bin/cam-event-detector && cam-event-detector --cam-id=test1 --video-source-url=/workspace/cam-event-detector/samples/cars1.mp4 --event-post-endpoint=http://dummy-event-receiver:3000/events

• Window references used for image debugging will be shown in your host computer (ex.: "window.IMShow(img)")

• The container may be running even on the cloud as the SSH connection will tunnel the X connection

• Performance profiling

  • go tool pprof -tree http://localhost:6060/debug/pprof/profile?seconds=5

Samples

• You can test this with video samples from http://github.com/flaviostutz/camera-samples

• Some live feeds

  • http://195.252.80.186:9000/mjpg/video.mjpg
  • http://50.73.9.194/mjpg/video.mjpg
  • http://83.14.13.226:81/mjpg/1/video.mjpg?camera=1&timestamp=1582556891831
  • http://209.206.162.229/mjpg/1/video.mjpg?camera=1&timestamp=1582557330589
  • http://87.57.111.162/mjpg/video.mjpg
  • http://83.14.13.226:81/mjpg/4/video.mjpg?camera=4&timestamp=1582561415901