Skip to content

Latest commit

 

History

History
191 lines (170 loc) · 7.95 KB

README.md

File metadata and controls

191 lines (170 loc) · 7.95 KB
Raw

Open the Way

This is a repository to play with OpenCV and yolo object detection model

Inspired by:

Requirements

  • Java 20
  • A default video camera (for demo purposes)
  • If using CUDA, you will need the appropriate GPU with the correct drivers
  • Docker, if you wish to run a local rtmp server

Quick start

First run

$ ./mvnw clean install

And execute with

$ ./mvnw exec:java

You should see an Canvas Frame appearing

YOLO object detection

In order to use object detection, download the official YOLOv7 files Once downloaded put them at the root of the project this way

 model/yolov7/config.cfg
 model/yolov7/names.txt
 model/yolov7/weights.weights

Then execute:

$ ./mvnw exec:java "-Dexec.args=--yolo-enabled --cuda-enabled"

If you wish to set your own path you can set it with the --yolo-path option

Read or write from a stream

You can read/write from a rtmp stream using respectively the option --input-type=stream --output-type=stream

In order to read from a rtmp server or to stream on it you can use this project: nginx-rtmp-docker

Start the project in a container this way:

docker run  -it --cpus 1 --memory 512m  -p 1935:1935 -p 8080:80 --rm alfg/nginx-rtmp

Then:

  • push streaming content to your server:
$ ./mvnw exec:java "-Dexec.args=--output-type=stream"
  • read that content, apply object detection and display it:
$ ./mvnw exec:java "-Dexec.args=--input-type=stream --yolo-enabled --cuda-enabled"

You should see the content streamed from your webcam

Full command usage

Usage: <main class> [--cuda-enabled] [--yolo-enabled]
                    [--confidence-threshold=<confidenceThreshold>]
                    [--gop=<gop>] [--in-address=<inputAddress>]
                    [--in-bitrate=<inputBitrate>] [--in-codec=<inputCodec>]
                    [--in-crf=<inputCrf>] [--in-format=<inputFormat>]
                    [--in-frame-rate=<inputFrameRate>]
                    [--in-height=<inputHeight>] [--in-preset=<inputPreset>]
                    [--in-tune=<inputTune>] [--in-type=<inputType>]
                    [--in-width=<inputWidth>] [--model-format=<modelFormat>]
                    [--nms-threshold=<nmsThreshold>]
                    [--out-address=<outputAddress>]
                    [--out-bitrate=<outputBitrate>] [--out-codec=<outputCodec>]
                    [--out-crf=<outputCrf>] [--out-format=<outputFormat>]
                    [--out-frame-rate=<outputFrameRate>]
                    [--out-height=<outputHeight>] [--out-preset=<outputPreset>]
                    [--out-tune=<outputTune>] [--out-type=<outputType>]
                    [--out-width=<outputWidth>] [--yolo-height=<yoloHeight>]
                    [--yolo-path=<modelPath>] [--yolo-width=<yoloWidth>]
      --confidence-threshold=<confidenceThreshold>
                            the confidence detection threshold
                            default: 0.4
      --cuda-enabled        enables cuda detection if present
                            fallbacks to CPU otherwise
                              Default: false
      --gop=<gop>           The output GOP (Group Of Pictures) size of the
                              stream
                              Default: 60
      --in-address=<inputAddress>
                            The input address to use when fetching from a URL
                            The URL depends on the forman (flv, mjpeg, ...) and
                              is optional if using --in-type=direct
                              Default: rtmp://localhost:1935/stream/hello
      --in-bitrate=<inputBitrate>
                            The input bitrate of the stream
                              Default: 2000000
      --in-codec=<inputCodec>
                            The input codec of the stream (default:
                              AV_CODEC_ID_H264)
      --in-crf=<inputCrf>   The input crf of the stream
                              Default: 28
      --in-format=<inputFormat>
                            The input format of the stream
                              Default: flv
      --in-frame-rate=<inputFrameRate>
                            The input frame rate of the stream
                              Default: 60
      --in-height=<inputHeight>
                            The input height of the stream
                              Default: 720
      --in-preset=<inputPreset>
                            The input preset of the stream
                              Default: ultrafast
      --in-tune=<inputTune> The input tune of the stream
                              Default: zerolatency
      --in-type=<inputType> The input type of the stream
                              Default: DIRECT
      --in-width=<inputWidth>
                            The input width of the stream
                              Default: 1280
      --model-format=<modelFormat>
                            enables yolo detection on the stream
                            default: DARKNET
                              Default: DARKNET
      --nms-threshold=<nmsThreshold>
                            the NMS (Non-maximum Suppression) threshold
                            more info: https://arxiv.org/abs/1705.02950
                              Default: 0.4
      --out-address=<outputAddress>
                            The output address to use when streaming on a URL
                            The URL depends on the format (flv, mjpeg, ...) and
                              is optional if using --out-type=direct
                              Default: rtmp://localhost:1935/stream/hello
      --out-bitrate=<outputBitrate>
                            The output bitrate of the stream
                              Default: 2000000
      --out-codec=<outputCodec>
                            The output codec of the stream (default)
      --out-crf=<outputCrf> The output crf of the stream
                              Default: 28
      --out-format=<outputFormat>
                            The output format of the stream
                              Default: flv
      --out-frame-rate=<outputFrameRate>
                            The output framerate of the stream
                              Default: 60
      --out-height=<outputHeight>
                            The output height of the stream
                              Default: 720
      --out-preset=<outputPreset>
                            The output preset of the stream
                              Default: ultrafast
      --out-tune=<outputTune>
                            The output tune of the stream
                              Default: zerolatency
      --out-type=<outputType>
                            The output type of the stream
                              Default: DIRECT
      --out-width=<outputWidth>
                            The output width of the stream
                              Default: 1280
      --yolo-enabled        enables yolo detection on the stream
                            default: false
                              Default: false
      --yolo-height=<yoloHeight>
                            the yolo image blob height used for detection
                              Default: 608
      --yolo-path=<modelPath>
                            path to the yolo files
                            the files within that folder must be named: config.
                              cfg, names.txt, weights.weights
                            eg: /path/to/yolov7 is the path that will contain:
                              /path/to/yolov7/config.cfg, /path/to/yolov7/names.
                              txt, /path/to/yolov7/weights.weights
      --yolo-width=<yoloWidth>
                            the yolo image blob width used for detection