Please follow these recommendations for proper fire and smoke detection:

  1. Use color cameras. With black and white cameras, the detection quality can be noticeably worse.
  2. The video resolution must be at least 640x360.
  3. FPS directly affects detection efficiency: in most cases, the default value (1 detection per 10 seconds) is sufficient to detect smoke or fire which develops in more than one minute.

  4. The minimal recognized fire/smoke area depends on the particular neural network used. For a standard neural network (see 화재 및 연기 검출 도구 설정), the fire/smoke area must occupy at least 10% of the frame.

    In some cases, when fire is expected to be clearly visible, the fire area may be sufficient at 1-3% of the FoV width/height.


  5. Fire/smoke should be visually separated from the background.

If you set a rectangular recognition area for the detection tool, the limitations apply to this area rather than to the entire frame (see 화재 및 연기 검출 도구 설정).


신경망 분석 작업을 위한 하드웨어 요구 사항