search

roflcoopter / viseron

Self-hosted, local only NVR and AI Computer Vision software. With features such as object detection, motion detection, face recognition and more, it gives you the power to keep an eye on your home, office or any other place you want to monitor.
MIT License
1.5k stars 154 forks source link

object_detector failure with darknet #740

Open csuther28 opened 2 months ago

csuther28 commented 2 months ago

I get the following failure once viseron started up in docker:

''' [2024-05-02 17:18:31] [DEBUG ] [viseron.components.nvr.nvr.camera_1] - Initializing NVR for camera Driveway [2024-05-02 17:18:31] [DEBUG ] [viseron.components.nvr.nvr.camera_1] - Waiting for first frame [2024-05-02 17:18:31] [DEBUG ] [viseron.components.ffmpeg.camera.camera_1] - Starting capture thread [2024-05-02 17:18:31] [DEBUG ] [viseron.components.ffmpeg.stream.camera_1] - FFmpeg decoder command: ffmpeg_camera_1 -hide_banner -loglevel error -avoid_negative_ts make_zero -fflags nobuffer -flags low_delay -strict experimental -fflags +genpts -use_wallclock_as_timestamps 1 -vsync 0 -timeout 5000000 -rtsp_transport tcp -i rtsp://:@@192.168.0.103:554/h264Preview_01_main -f segment -segment_time 5 -reset_timestamps 1 -strftime 1 -c:v copy -c:a aac /segments/camera_1/%Y%m%d%H%M%S.mp4 -vf fps=2.0 -f rawvideo -pix_fmt nv12 pipe:1 [2024-05-02 17:18:31] [INFO ] [viseron.components.nvr.nvr.camera_1] - NVR for camera Driveway initialized [2024-05-02 17:18:31] [INFO ] [viseron.components] - Setup of domain nvr for component nvr with identifier camera_1 took 0.0 seconds [2024-05-02 17:18:31] [INFO ] [viseron.core] - Viseron initialized in 4.4 seconds [2024-05-02 17:18:36] [DEBUG ] [viseron.components.nvr.nvr.camera_1] - First frame received [2024-05-02 17:18:39] [ERROR ] [viseron.components.nvr.nvr.camera_1] - Failed to retrieve result for object_detector [2024-05-02 17:18:39] [DEBUG ] [viseron.components.nvr.nvr.camera_1] - Frame is 2.531282901763916 seconds old. Discarding [2024-05-02 17:18:39] [DEBUG ] [viseron.components.nvr.nvr.camera_1] - Frame is 2.0378434658050537 seconds old. Discarding [2024-05-02 17:18:39] [DEBUG ] [viseron.components.nvr.nvr.camera_1] - Frame is 1.4673566818237305 seconds old. Discarding [2024-05-02 17:18:39] [DEBUG ] [viseron.components.nvr.nvr.camera_1] - Frame is 1.0376908779144287 seconds old. Discarding [2024-05-02 17:18:42] [ERROR ] [viseron.components.nvr.nvr.camera_1] - Failed to retrieve result for object_detector [2024-05-02 17:18:42] [DEBUG ] [viseron.components.nvr.nvr.camera_1] - Frame is 3.105250120162964 seconds old. Discarding [2024-05-02 17:18:42] [DEBUG ] [viseron.components.nvr.nvr.camera_1] - Frame is 2.5575098991394043 seconds old. Discarding '''

The following is my config: ''' ffmpeg: camera: camera_1: # This value has to be unique across all cameras name: Driveway host: 192.168.0.103 port: 554 path: /h264Preview_01_main protocol: rtsp width: 2560 height: 1920 fps: 15 audio_codec: aac username: !secret ovif_user password: !secret ovif_pwd recorder: filename_pattern: '%H%M%S'

darknet: object_detector: cameras: camera_1: # Attach detector to the configured camera_1 above fps: 2 scan_on_motion_only: false # false = Scan for objects even when there is no motion log_all_objects: false labels:

Any idea what I'm doing wrong?

The motion recording is working great though.

csuther28 commented 2 months ago

The following observation via logs the next day for above camera:

''' [2024-05-03 10:25:48] [DEBUG ] [viseron.components.nvr.nvr.camera_1] - Frame is 1.148996114730835 seconds old. Discarding [2024-05-03 10:25:48] [DEBUG ] [viseron.components.mog2.motion_detector.camera_1] - Max motion area: 0.00218 [2024-05-03 10:25:51] [ERROR ] [viseron.components.nvr.nvr.camera_1] - Failed to retrieve result for object_detector [2024-05-03 10:25:51] [DEBUG ] [viseron.components.nvr.nvr.camera_1] - Frame is 3.203981876373291 seconds old. Discarding [2024-05-03 10:25:51] [DEBUG ] [viseron.components.nvr.nvr.camera_1] - Frame is 2.7307252883911133 seconds old. Discarding [2024-05-03 10:25:51] [DEBUG ] [viseron.components.nvr.nvr.camera_1] - Frame is 2.2098772525787354 seconds old. Discarding [2024-05-03 10:25:51] [DEBUG ] [viseron.components.nvr.nvr.camera_1] - Frame is 1.57594895362854 seconds old. Discarding [2024-05-03 10:25:51] [DEBUG ] [viseron.components.nvr.nvr.camera_1] - Frame is 1.201096534729004 seconds old. Discarding [2024-05-03 10:25:51] [DEBUG ] [viseron.components.mog2.motion_detector.camera_1] - Max motion area: 0.00161 [2024-05-03 10:25:54] [DEBUG ] [viseron.components.darknet.object_detector.camera_1] - Objects: [{'label': 'person', 'confidence': 0.869, 'rel_width': 0.092, 'rel_height': 0.334, 'rel_x1': 0.375, 'rel_y1': 0.061, 'rel_x2': 0.467, 'rel_y2': 0.395}] [2024-05-03 10:25:54] [DEBUG ] [viseron.components.darknet.object_detector.camera_1] - Frame is 3.518185615539551 seconds old. Discarding [2024-05-03 10:25:54] [DEBUG ] [viseron.components.nvr.nvr.camera_1] - Frame is 3.085675001144409 seconds old. Discarding [2024-05-03 10:25:54] [DEBUG ] [viseron.components.nvr.nvr.camera_1] - Frame is 2.3212411403656006 seconds old. Discarding '''

From this I gather that the OBJECT DETECTOR is working and detected a person within the area. But still with the issue 'Failed to retrieve result for object_detector'

roflcoopter commented 1 month ago

Hmm strange, seems that inference is too slow. Is your CPU maxed out while Viseron is running?

csuther28 commented 1 month ago

When I started off, all seems ok...except for the mentioned error alert....at some time did I detect a slowness on server, on closer look, the CPUs were maxed out...I then disabled the object_detector to run continues without motion detector...after this did CPU normalize.......but this alert still popping up.

roflcoopter commented 1 month ago

Does it get any better if you set fps: 1 for the object_detector?