deepstream-python
deepstream-python copied to clipboard
ml6team
Customizing multi camera streaming pipeline
Hi, Thanks for your great work. I have a query regarding the multi-camera stream pipeline. In my case, I have a config file having 4 video sources in which I am applying object detection using Yolov5. Later after detecting objects, I have to calculate distances and had to do other customization on bounding boxes and acquired frames. How can I acquire this data in my python script? Any suggestion, please? @Joxis
Hi @sharoseali! Glad you like the repository 🙂
Could you provide a bit more information on what exactly you want to achieve? Do you need the timestamps at which objects were detected? With the current implementation you can already access frames and detections via a probe.
Here is the working config file that I use to get a video stream from 4 video files. Now I want to get these frames and corresponding detections for customization and display again them on 4 screens.
[application]
enable-perf-measurement=1
perf-measurement-interval-sec=5
[tiled-display]
enable=1
rows=2
columns=2
width=1280
height=720
gpu-id=0
nvbuf-memory-type=0
[source0]
enable=1
#Type - 1=CameraV4L2 2=URI 3=MultiURI
type=1
camera-width=640
camera-height=480
camera-fps-n=30
camera-fps-d=1
camera-v4l2-dev-node=4
[source1]
enable=1
#Type - 1=CameraV4L2 2=URI 3=MultiURI
type=1
camera-width=640
camera-height=480
camera-fps-n=30
camera-fps-d=1
camera-v4l2-dev-node=2
[source2]
enable=1
#Type - 1=CameraV4L2 2=URI 3=MultiURI
type=1
camera-width=640
camera-height=480
camera-fps-n=30
camera-fps-d=1
camera-v4l2-dev-node=4
[source3]
enable=1
type=3
uri=file:/home/rpt/Udentify/keas_trt_code/inputs/3.mp4
num-sources=1
gpu-id=0
cudadec-memtype=0
[sink0]
enable=1
type=2
sync=0
gpu-id=0
nvbuf-memory-type=0
[osd]
enable=1
gpu-id=0
border-width=5
text-size=15
text-color=1;1;1;1;
text-bg-color=0.3;0.3;0.3;1
font=Serif
show-clock=0
clock-x-offset=800
clock-y-offset=820
clock-text-size=12
clock-color=1;0;0;0
nvbuf-memory-type=0
[streammux]
gpu-id=0
live-source=0
batch-size=4
batched-push-timeout=40000
width=1920
height=1080
enable-padding=0
nvbuf-memory-type=0
[primary-gie]
enable=1
gpu-id=0
gie-unique-id=1
nvbuf-memory-type=0
config-file=config_infer_primary_yoloV5.txt
[tests]
file-loop=0
Could you provide a bit more information on what exactly you want to achieve?
I want to calculate distances and redraw some data values based on B-boxes values.
@Joxis Any response, please? Explaining a bit more I wanna use my YOLOV5 config file with python to extract boxes using deep stream. Meanwhile, I need two things to customize my detector output. 1 -Camera frame (I will draw a circle using open CV) and bounding boxes (x1 y1, x2, y2) to plot them differently. Sorry, if you don't understand my last comment.
The current implementation does not support multiple video streams yet. Most code is already in place to make this possible but some final adjustments are needed. Did you already manage to get the boilerplate working with multiple streams? Or did you only use the standard sample application?
Meanwhile, I am still trying to manage multistream in the current code. I want to ask which specific things I have to edit in app/pipeline.py. Also the current implementation is with tracker but I don't need it for now. So how should I replace my pipeline without a tracker? Thanks
The _create_source_bin function should already create a bin supporting multiple streams. You will, however, have to adjust how it is called (the index parameter is currently not used). An example can be found here.
With respect to the tracker, the easiest method is to just use the IOU Tracker. It is fast and should not impact performance too much. If you really want to remove the tracker you would have to remove this line. You can also subclass the base pipeline and overwrite the _create_tracker function to perform nothing.
HI I almost manage to arrange the code for multi streaming and is almost nearly to work (So I deleted my last comment).
But meanwhile running the code after changes i came across the issue in live camera streaming. As I want to read stream from USB cams that I connected with my Jetson NX kit, so I provide camera ids in this way:
['/dev/video3', '/dev/video2', '1.mp4', '2.mp4'] for 4 sources.
I got the following error in gstreamer pipeline. I dont know whats the reason behind that . if someone knows please share. Here are my output logs:
python3 run.py
INFO:app.Custom_pipeline.AnonymizationPipeline:Playing from URI None
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating Pipeline
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating Stream mux
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating Source bin
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating URI decode bin
INFO:app.Custom_pipeline.AnonymizationPipeline:Linking elements in the Pipeline: stream-muxer -> source-bin-00
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating Source bin
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating URI decode bin
INFO:app.Custom_pipeline.AnonymizationPipeline:Linking elements in the Pipeline: stream-muxer -> source-bin-00 -> source-bin-01
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating Source bin
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating URI decode bin
INFO:app.Custom_pipeline.AnonymizationPipeline:Linking elements in the Pipeline: stream-muxer -> source-bin-00 -> source-bin-01 -> source-bin-02
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating Source bin
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating URI decode bin
INFO:app.Custom_pipeline.AnonymizationPipeline:Linking elements in the Pipeline: stream-muxer -> source-bin-00 -> source-bin-01 -> source-bin-02 -> source-bin-03
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating PGIE
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating Converter 1
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating Caps filter 1
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating Tiler
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating Converter 2
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating Queue 1
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating Converter 3
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating Caps filter 2
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating Encoder
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating Parser
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating Container
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating Sink
INFO:app.Custom_pipeline.AnonymizationPipeline:Starting pipeline
WARNING: [TRT]: Using an engine plan file across different models of devices is not recommended and is likely to affect performance or even cause errors.
Deserialize yoloLayer plugin: yolo
0:00:08.110955645 7949 0x301108d0 INFO nvinfer gstnvinfer.cpp:646:gst_nvinfer_logger:<primary-inference> NvDsInferContext[UID 1]: Info from NvDsInferContextImpl::deserializeEngineAndBackend() <nvdsinfer_context_impl.cpp:1909> [UID = 1]: deserialized trt engine from :/home/rpt/Udentify/DeepStream-Yolo/deepstream-python/deepstream/configs/model_b1_gpu0_fp32.engine
INFO: [Implicit Engine Info]: layers num: 5
0 INPUT kFLOAT data 3x320x320
1 OUTPUT kFLOAT num_detections 1
2 OUTPUT kFLOAT detection_boxes 6375x4
3 OUTPUT kFLOAT detection_scores 6375
4 OUTPUT kFLOAT detection_classes 6375
0:00:08.183567375 7949 0x301108d0 INFO nvinfer gstnvinfer.cpp:646:gst_nvinfer_logger:<primary-inference> NvDsInferContext[UID 1]: Info from NvDsInferContextImpl::generateBackendContext() <nvdsinfer_context_impl.cpp:2012> [UID = 1]: Use deserialized engine model: /home/rpt/Udentify/DeepStream-Yolo/deepstream-python/deepstream/configs/model_b1_gpu0_fp32.engine
0:00:08.195152594 7949 0x301108d0 INFO nvinfer gstnvinfer_impl.cpp:328:notifyLoadModelStatus:<primary-inference> [UID 1]: Load new model:/home/rpt/Udentify/DeepStream-Yolo/deepstream-python/deepstream/app/../configs/deepstream_app_config_4_cam.txt sucessfully
Error: gst-resource-error-quark: Invalid URI "/dev/video3". (3): gsturidecodebin.c(1383): gen_source_element (): /GstPipeline:pipeline0/GstBin:source-bin-00/GstURIDecodeBin:uri-decode-bin
INFO:app.Custom_pipeline.AnonymizationPipeline:Exiting pipeline
Available camera ids showed by this command v4l2-ctl --list-devices is below
NVIDIA Tegra Video Input Device (platform:tegra-camrtc-ca):
/dev/media0
Webcam C170: Webcam C170 (usb-3610000.xhci-2.1):
/dev/video0
/dev/video1
/dev/media1
Webcam C170: Webcam C170 (usb-3610000.xhci-2.2):
/dev/video2
/dev/video3
/dev/media2
One thing more I have changed here in _create_mp4_sink_bin function is
mp4_sink_bin.add_pad(Gst.GhostPad("sink", nvvidconv3.get_static_pad("sink"))) to
bin_pad = mp4_sink_bin.add_pad(Gst.GhostPad.new_no_target("src", Gst.PadDirection.SRC).
This is because i was getting error :
INFO:app.Custom_pipeline.AnonymizationPipeline:Creating Sink
Traceback (most recent call last):
File "run.py", line 11, in <module>
run_anonymization_pipeline(args.list)
File "/home/rpt/Udentify/DeepStream-Yolo/deepstream-python/deepstream/app/core.py", line 32, in run_anonymization_pipeline
pipeline = AnonymizationPipeline(
File "/home/rpt/Udentify/DeepStream-Yolo/deepstream-python/deepstream/app/pipelines/anonymization.py", line 19, in __init__
super().__init__(*args, **kwargs)
File "/home/rpt/Udentify/DeepStream-Yolo/deepstream-python/deepstream/app/Custom_pipeline.py", line 107, in __init__
self._create_elements()
File "/home/rpt/Udentify/DeepStream-Yolo/deepstream-python/deepstream/app/Custom_pipeline.py", line 364, in _create_elements
self.sink_bin = self._create_mp4_sink_bin()
File "/home/rpt/Udentify/DeepStream-Yolo/deepstream-python/deepstream/app/Custom_pipeline.py", line 286, in _create_mp4_sink_bin
mp4_sink_bin.add_pad(Gst.GhostPad("sink", nvvidconv3.get_static_pad("sink")))
TypeError: GObject.__init__() takes exactly 0 arguments (2 given)
So where I am exactly wrong? Any guidance...