justinburger / dinetta

IP Camera Object Detection, Training and Alerting

Dinetta

IP Camera Object Detection, Training and Alerting

Project Objectives

Provide a simple to use object (person) dectection, training and alerting system that will run on a raspberry pi, pulling images from multiple IP Cameras for near realtime alerting.

Functionality Goals

• User Defined Settings:

  • Alerting
    • How often to see consider a detection as the same event.
    • Where to email on detection (differnt ones for differnt detections)
    • What % likleyhood to alert on
    • How many Different camera sources to alert on.
    • Different settings per time of day.
    • Different alerting schedules for differnt types (immedate, per day/per week)
    • How many training images to store before rotating.
  • Off device storage options (S3)

• Web based UI to select objects on still frame images used for training object detection.

Project Componets

Image Capture Service

./manage image_capture_service [--all]

The Image Capture Service provides the functionality to pull still images from defined IP Cameras via RTSP.

Image Trainer

Provides the front/backend needed to work thru stored images while training.