New Custom Compontent - Image Processing - Object Detection - DOODS

Share your Projects!
635

Oh, and the HA Green is pi/arm 64 based?
SoC Rockchip RK3566 SoC with quad-core Arm Cortex-A55 CPU
CPU Frequency: 1.8 GHz

Any other model recommendations to try and increase efficiency, my primary use case is to recognize people, animals are a bonus… My use case triggers a single frame/pic after motion on a set of cameras and passes it off to DOODS to recognize people. I’d love if I could get that recognition sub-second, right now it seems to range from 4-9 seconds in my testing, although CPU and memory load seem fine.

636

Yeah, that’s ARM… cheap desktops can be had on Ebay… Intel i5-6600 is a good one.

637

Thanks. Any recommendations on models to target people, that are efficient and can work with DOODS on ARM? I just don’t know where to get started with other models. Thanks.

638

Just look on Ebay or whatever you use to find a cheap desktop with 16GB of ram and and Intel i5. Dell, HP and Lenovo all make them. Figure around 100$. I don’t know what the current cheap desktops are. Basically as companies replace them, there will be zillions of the same model. Maybe like this: Dell OptiPlex Mini Pc Desktop Computer Intel Core i5 16GB RAM 1TB HD Windows 10 | eBay

1 Like
639

@snowzach , is it possible to set up multiple areas for the same object? if I add the same label twice, but with a different area, it only detects the second one. I tried removing the second zone and it gets detected in the first one ok. As a workaround, I set it to detect ,sheep" on the second area. It has a power confidence, but it works.

image_processing:
  - platform: doods
    url: http://10.10.10.10:8098
    detector: pytorch
    scan_interval: 2
    source:
      - entity_id: camera.10_10_10_209
    file_out: /config/www/doods.jpg
    confidence: 20
    area:
      top: 0
      bottom: 1
      left: 0
      right: 1
      covers: true
    labels:
      - name: dog
        confidence: 20
        area:
          top: 0.2
          left: 0.02
          bottom: 0.53
          right: 0.8
          covers: false
      - name: dog
        confidence: 20
        area:
          top: 0.53
          left: 0.02
          bottom: 0.98
          right: 0.38
          covers: false