Low quality video streaming

[tunahsu]

Why is the video streaming quality poor even though I've set the resolution to the highest resolution of the webcam in the media_stream_constraints? The quality is still worse than directly opening the camera using Windows.

Here is my code:

ctx = webrtc_streamer(
    key='stream-detection',
    video_processor_factory=VideoProcessor,
    media_stream_constraints={
        'video': {
            'width': 1920
            },
        'audio': False
    },
    rtc_configuration={
        'iceServers': [{'urls': ['stun:stun.l.google.com:19302']}]
    },
)

[tunahsu]

I noticed that it takes some time for the frame size to gradually increase, from 640x360 to 960x540 to 1280x720 to 1920x1080. I know that WebRTC adjusts the frame size adaptively based on network speed. Is there any way to fix it at 1920x1080 from the beginning?

[ccl-private]

I also encountered this problem. 1920x1080 video streaming quality is poor. Is your problem solved?

[tunahsu]

@ccl-private It seems that manual adjustments in the package are not possible... The author mentioned it here.

[ccl-private]

I can understand it, but the difference in image clarity between local playback and webrtc playback is too big.

[ccl-private]

OK, I tried https://github.com/aiortc/aiortc/tree/main/examples/webcam. Also Low quality for l080p.

[achaosss]

I have the same question

[ccl-private]

@tunahsu @achaosss @whitphx Hei guys, I found an article that may help.https://www.freelancer.com/contest/aiortc-example-with-hardware-accelerated-codec-2035986, but it’s a bit difficult to solve it with my own ability.

[ccl-private]

import numpy as np
import subprocess
import re

# 定义 RTSP URL
rtsp_url = "rtsp://192.168.2.13:554/cam/realmonitor?channel=1&subtype=0"

# 使用 ffprobe 获取视频分辨率
def get_video_resolution(rtsp_url):
    command = [
        'ffprobe',
        '-v', 'error',
        '-select_streams', 'v:0',
        '-show_entries', 'stream=width,height',
        '-of', 'csv=p=0',
        rtsp_url
    ]

    result = subprocess.run(command, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
    if result.returncode == 0:
        resolution = result.stdout.decode('utf-8').strip().split(',')
        width, height = int(resolution[0]), int(resolution[1])
        return width, height
    else:
        print("Error:", result.stderr.decode('utf-8'))
        return None

# 获取视频分辨率
resolution = get_video_resolution(rtsp_url)
if resolution:
    width, height = resolution
    print("视频分辨率:", resolution)

    # # 启动 FFmpeg 子进程
    # ffmpeg_command = [
    #     'ffmpeg',
    #     # '-loglevel', 'debug',  # 启用调试日志
    #     '-i', rtsp_url,
    #     '-f', 'rawvideo',
    #     '-pix_fmt', 'bgr24',
    #     '-'
    # ]
    ffmpeg_command = [
        'ffmpeg',
        '-hwaccel', 'vaapi',
        '-vaapi_device', '/dev/dri/renderD128',
        '-i', rtsp_url,
        '-f', 'rawvideo',
        '-pix_fmt', 'bgr24',
        '-'
    ]

    process = subprocess.Popen(ffmpeg_command, stdout=subprocess.PIPE, stderr=subprocess.PIPE)

    cv2.namedWindow("Video Stream", cv2.WINDOW_NORMAL)

    while True:
        # 根据分辨率读取数据
        raw_frame = process.stdout.read(width * height * 3)  # BGR格式每个像素3字节
        if not raw_frame:
            break

        # stderr_line = process.stderr.readline()
        # if not stderr_line:
        #     break
        # print(stderr_line.strip())

        # 将原始数据转换为 NumPy 数组
        frame = np.frombuffer(raw_frame, np.uint8)
        frame = frame.reshape((height, width, 3))  # 适应 BGR格式

        # 显示图像
        cv2.imshow("Video Stream", frame)
        if cv2.waitKey(1) & 0xFF == ord('q'):
            break

    # 清理
    process.stdout.close()
    process.stderr.close()
    process.wait()
    cv2.destroyAllWindows()
else:
    print("无法获取视频分辨率。")

I can call hardware decoding directly with ffmpeg to extract the RTSP stream to the frame variable in real time, but I don't know how to pass it to streamlit to display it on the interface

[ccl-private]

Hello everyone, I simply wrote a project to pull 2560*1440 streams in real time and display them on streamlit web. https://github.com/ccl-private/streamlit_rtsp/tree/main
The project is still very simple. It uses streaming ffmpeg hardware to pull the stream (Intel core display). The frame callback has not been written yet. Of course, you can also modify it directly in the source code.

@tunahsu @achaosss