Increase detection accuracy

[tuanha2000vn]

What is the best setting to increase the accuracy, my camera keep reporting person instead of dog and I've set the threadhold pretty high

"camera_2": {
      "best_image_timeout": 60,
      "clips": {
        "enabled": true,
        "objects": [
          "person",
          "car",
          "dog",
          "cat"
        ],
        "post_capture": 5,
        "pre_capture": 5,
        "required_zones": [],
        "retain": {
          "default": 3,
          "objects": {}
        }
      },
      "detect": {
        "enabled": true,
        "max_disappeared": 20
      },
      "ffmpeg_cmds": [
        {
          "cmd": "ffmpeg -hide_banner -loglevel warning -hwaccel vaapi -hwaccel_device /dev/dri/renderD128 -hwaccel_output_format yuv420p -avoid_negative_ts make_zero -fflags +genpts+discardcorrupt -rtsp_transport tcp -stimeout 5000000 -use_wallclock_as_timestamps 1 -i rtsp://admin:[email protected]:554/Streaming/Channels/101 -f segment -segment_time 10 -segment_format mp4 -reset_timestamps 1 -strftime 1 -c copy -an /tmp/cache/camera_2-%Y%m%d%H%M%S.mp4",
          "roles": [
            "clips",
            "rtmp"
          ]
        },
        {
          "cmd": "ffmpeg -hide_banner -loglevel warning -hwaccel vaapi -hwaccel_device /dev/dri/renderD128 -hwaccel_output_format yuv420p -avoid_negative_ts make_zero -fflags +genpts+discardcorrupt -rtsp_transport tcp -stimeout 5000000 -use_wallclock_as_timestamps 1 -i rtsp://admin:[email protected]:554/Streaming/Channels/103 -r 4 -f rawvideo -pix_fmt yuv420p pipe:",
          "roles": [
            "detect"
          ]
        }
      ],
      "fps": 4,
      "frame_shape": [
        720,
        1280
      ],
      "height": 720,
      "motion": {
        "contour_area": 100,
        "delta_alpha": 0.2,
        "frame_alpha": 0.2,
        "frame_height": 120,
        "mask": [
          "1084,59,914,21,86,30,0,169,0,0,1280,0,1280,147"
        ],
        "threshold": 25
      },
      "mqtt": {
        "bounding_box": true,
        "crop": true,
        "enabled": true,
        "height": 270,
        "required_zones": [],
        "timestamp": true
      },
      "name": "camera_2",
      "objects": {
        "filters": {
          "car": {
            "mask": [
              "1084,59,914,21,86,30,0,169,0,0,1280,0,1280,147"
            ],
            "max_area": 24000000,
            "min_area": 0,
            "min_score": 0.5,
            "threshold": 0.75
          },
          "cat": {
            "mask": [
              "1084,59,914,21,86,30,0,169,0,0,1280,0,1280,147"
            ],
            "max_area": 24000000,
            "min_area": 0,
            "min_score": 0.5,
            "threshold": 0.7
          },
          "dog": {
            "mask": [
              "1084,59,914,21,86,30,0,169,0,0,1280,0,1280,147"
            ],
            "max_area": 24000000,
            "min_area": 0,
            "min_score": 0.5,
            "threshold": 0.7
          },
          "person": {
            "mask": [
              "1084,59,914,21,86,30,0,169,0,0,1280,0,1280,147"
            ],
            "max_area": 24000000,
            "min_area": 0,
            "min_score": 0.5,
            "threshold": 0.75
          }
        },
        "mask": [
          "1084,59,914,21,86,30,0,169,0,0,1280,0,1280,147"
        ],
        "track": [
          "person",
          "car",
          "dog",
          "cat"
        ]
      },
      "record": {
        "enabled": false,
        "retain_days": 30
      },
      "rtmp": {
        "enabled": false
      },
      "snapshots": {
        "bounding_box": true,
        "crop": false,
        "enabled": true,
        "height": null,
        "required_zones": [],
        "retain": {
          "default": 10,
          "objects": {}
        },
        "timestamp": false
      },
      "width": 1280,
      "zones": {}
    },

[anwik]

I think you have to set a higher value for "min_area" in your object filter. You might have to setup extra zones as well since a person can have different values in different areas of the camera view.

I recently started using Frigate though so I'm no expert!

[DrSpaldo]

I am also getting a similar issue with a toy car triggering a person detection and was wondering how I could stop it