使用 NODE.JS 和 html5 实现低延迟（50 毫秒）视频流

Question

目标：

我正在构建一个 FPV 机器人，我想通过本地 Wi-Fi 连接使用网络浏览器来控制它。

我使用的是带有 Raspbian Stretch 的树莓派 3B+。我建造了自己的电机控制和电源调节器帽子。

经过大量研究测试，我决定使用node.JS作为http服务器和socket.io来提供与我的机器人的低延迟双向通信。该堆栈实现了大约 7 毫秒的延迟。

问题：

我需要将低延迟视频从连接到 RPI 的 USB 摄像头传输到浏览器。我的目标是在 10FPS 下实现至少 640x480 分辨率，并具有 50 毫秒或更好的延迟。我很高兴牺牲视觉逼真度来获得机器人更快的响应。

如果可能的话我想以UDP方式进行流传输以提高流的可靠性。

如果可能的话，我想传输现代网络浏览器可以本地解码的视频。我想使用 H264 编解码器和 HTML5 视频标签。 如果没有其他选择，我可以回退到使用 javascript 播放器。

我尝试过：

我做了广泛的研究并尝试了很多工具。

除此之外，我尝试了 VLC、mjpg Streamer、gstreamer 和 raspivid。有几次我访问了网络浏览器可以查看的流，但在 320x240 分辨率下最多延迟 700 毫秒。离我的目标还很远。

目前我正在研究 WebRTC 解决方案。

问题：

我想要 NODE.JS 包或其他解决方案的建议，以提供 UDP H264 视频流，该视频流可以由 HTML5 视频标签解码，目标延迟为 50 毫秒。

谢谢

更新：

感谢您的回答！我会不断更新这个问题，一旦解决我会发布解决方案。

推送单个帧

我尝试了一种不同的方法，通过 websocket 推送单个 200KB 640x480 jpg 帧，延迟约为 190 毫秒。通过重用对象，我可能可以做得更好，但我现在暂时搁置这个尝试。

更新2：

在研究 WebRTC 时，我发现了一个看起来很简单的堆栈。 服务器端使用V4L2作为驱动程序，FFMPEG在本地转码为带有TS封装的MPEG1 http流，node js将流翻转为websocket。 客户端有一个 JavaScript 代码可以解码 MPEG1 TS 流并将画布对象绘制到 HTML 页面中。

它实现了 640x480@20FPS，延迟为 240 毫秒。 对于 MVP 来说已经足够好了，但我会继续努力争取它。 答案中的代码。

Answer 1

我从这里改编了代码并将其与http服务器和socket.io控件集成： https://github.com/phoboslab/jsmpeg

服务器：

V4L2 -> FFMPEG (MPEG1 TS) -> NODE HTTP 服务器 -> NODE Websocket 广播

客户：

Websocket -> Javascript（解码 MPEG1 TS 并绘制到 html 画布）-> Html Canvas

该堆栈实现了 640x480@20FPS，延迟为 240 毫秒。虽然距离我的目标还很远，但作为 MVP 已经足够了。两个方向的控制都有 7ms 的延迟，非常好。

该堆栈受到转码和解码阶段的阻碍，并且 RPI 变得非常热。通过 websocket 传输原始数据看起来不错，我将在未来分析每个步骤的延迟。

执行：

pi@MazeRunner:~ $ node node.js &
pi@MazeRunner:~ $ ffmpeg -f v4l2 -framerate 20 -video_size 640x480 -i /dev/video0 -f mpegts -codec:v mpeg1video -s 640x480 -b:v 600k -bf 0 http://localhost:8080/mystream

服务器端NODE.JS

//operating system library. Used to get local IP address
var os = require("os");
//file system library. Used to load file stored inside back end server (https://nodejs.org/api/fs.html)
var fs = require("fs");
//http system library. Handles basic html requests
var http = require("http").createServer(http_handler);
//url library. Used to process html url requests
var url = require("url");
//Websocket
var io = require("socket.io")(http);
//Websocket used to stream video
var websocket = require("ws");

//-----------------------------------------------------------------------------------
//	CONFIGURATION
//-----------------------------------------------------------------------------------

//Port the server will listen to
var server_port = 8080;
var websocket_stream_port = 8082;
//Path of the http and css files for the http server
var file_index_name = "index.html";
var file_css_name = "style.css";
var file_jsplayer_name = "jsmpeg.min.js";
//Http and css files loaded into memory for fast access
var file_index;
var file_css;
var file_jsplayer;
//Name of the local video stream
var stream_name = "mystream";

//-----------------------------------------------------------------------------------
//	DETECT SERVER OWN IP
//-----------------------------------------------------------------------------------

//If just one interface, store the server IP Here
var server_ip;
//Get local IP address of the server
//https://stackoverflow.com/questions/3653065/get-local-ip-address-in-node-js
var ifaces = os.networkInterfaces();

Object.keys(ifaces).forEach
(
	function (ifname)
	{
		var alias = 0;

		ifaces[ifname].forEach
		(
			function (iface)
			{
				if ('IPv4' !== iface.family || iface.internal !== false)
				{
				  // skip over internal (i.e. 127.0.0.1) and non-ipv4 addresses
				  return;
				}

				if (alias >= 1)
				{
					// this single interface has multiple ipv4 addresses
					console.log('INFO: Server interface ' +alias +' - ' + ifname + ':' + alias, iface.address);
				}
				else
				{
					server_ip = iface.address;
					// this interface has only one ipv4 adress
					console.log('INFO: Server interface - ' +ifname, iface.address);
				}
				++alias;
			}
		);
	}
);

//-----------------------------------------------------------------------------------
//	HTTP SERVER
//-----------------------------------------------------------------------------------
//	Fetch and serves local files to client

//Create http server and listen to the given port
http.listen
(
	server_port,
	function( )
	{
		console.log('INFO: ' +server_ip +' listening to html requests on port ' +server_port);
		//Pre-load http, css and js files into memory to improve http request latency
		file_index = load_file( file_index_name );
		file_css = load_file( file_css_name );
		file_jsplayer = load_file( file_jsplayer_name );
	}
);

//-----------------------------------------------------------------------------------
//	HTTP REQUESTS HANDLER
//-----------------------------------------------------------------------------------
//	Answer to client http requests. Serve http, css and js files

function http_handler(req, res)
{
	//If client asks for root
	if (req.url == '/')
	{
		//Request main page
		res.writeHead( 200, {"Content-Type": detect_content(file_index_name),"Content-Length":file_index.length} );
		res.write(file_index);
		res.end();

		console.log("INFO: Serving file: " +req.url);
	}
	//If client asks for css file
	else if (req.url == ("/" +file_css_name))
	{
		//Request main page
		res.writeHead( 200, {"Content-Type": detect_content(file_css_name),"Content-Length" :file_css.length} );
		res.write(file_css);
		res.end();

		console.log("INFO: Serving file: " +req.url);
	}
	//If client asks for css file
	else if (req.url == ("/" +file_jsplayer_name))
	{
		//Request main page
		res.writeHead( 200, {"Content-Type": detect_content(file_jsplayer_name),"Content-Length" :file_jsplayer.length} );
		res.write(file_jsplayer);
		res.end();

		console.log("INFO: Serving file: " +req.url);
	}
	//Listening to the port the stream from ffmpeg will flow into
	else if (req.url = "/mystream")
	{
		res.connection.setTimeout(0);

		console.log( "Stream Connected: " +req.socket.remoteAddress + ":" +req.socket.remotePort );

		req.on
		(
			"data",
			function(data)
			{
				streaming_websocket.broadcast(data);
				/*
				if (req.socket.recording)
				{
					req.socket.recording.write(data);
				}
				*/
				//console.log("broadcast: ", data.length);
			}
		);

		req.on
		(
			"end",
			function()
			{
				console.log("local stream has ended");
				if (req.socket.recording)
				{
					req.socket.recording.close();
				}
			}
		);

	}
	//If client asks for an unhandled path
	else
	{
		res.end();
		console.log("ERR: Invalid file request" +req.url);
	}
}

//-----------------------------------------------------------------------------------
//	WEBSOCKET SERVER: CONTROL/FEEDBACK REQUESTS
//-----------------------------------------------------------------------------------
//	Handle websocket connection to the client

io.on
(
	"connection",
	function (socket)
	{
		console.log("connecting...");

		socket.emit("welcome", { payload: "Server says hello" });

		//Periodically send the current server time to the client in string form
		setInterval
		(
			function()
			{
				socket.emit("server_time", { server_time: get_server_time() });
			},
			//Send every 333ms
			333
		);

		socket.on
		(
			"myclick",
			function (data)
			{
				timestamp_ms = get_timestamp_ms();
				socket.emit("profile_ping", { timestamp: timestamp_ms });
				console.log("button event: " +" client says: " +data.payload);
			}
		);

		//"ArrowLeft"
		socket.on
		(
			"keyboard",
			function (data)
			{
				timestamp_ms = get_timestamp_ms();
				socket.emit("profile_ping", { timestamp: timestamp_ms });
				console.log("keyboard event: " +" client says: " +data.payload);
			}
		);

		//profile packets from the client are answer that allows to compute roundway trip time
		socket.on
		(
			"profile_pong",
			function (data)
			{
				timestamp_ms_pong = get_timestamp_ms();
				timestamp_ms_ping = data.timestamp;
				console.log("Pong received. Round trip time[ms]: " +(timestamp_ms_pong -timestamp_ms_ping));
			}
		);
	}
);

//-----------------------------------------------------------------------------------
//	WEBSOCKET SERVER: STREAMING VIDEO
//-----------------------------------------------------------------------------------

// Websocket Server
var streaming_websocket = new websocket.Server({port: websocket_stream_port, perMessageDeflate: false});

streaming_websocket.connectionCount = 0;

streaming_websocket.on
(
	"connection",
	function(socket, upgradeReq)
	{
		streaming_websocket.connectionCount++;
		console.log
		(
			'New websocket Connection: ',
			(upgradeReq || socket.upgradeReq).socket.remoteAddress,
			(upgradeReq || socket.upgradeReq).headers['user-agent'],
			'('+streaming_websocket.connectionCount+" total)"
		);

		socket.on
		(
			'close',
			function(code, message)
			{
				streaming_websocket.connectionCount--;
				console.log('Disconnected websocket ('+streaming_websocket.connectionCount+' total)');
			}
		);
	}
);

streaming_websocket.broadcast = function(data)
{
	streaming_websocket.clients.forEach
	(
		function each(client)
		{
			if (client.readyState === websocket.OPEN)
			{
				client.send(data);
			}
		}
	);
};


//-----------------------------------------------------------------------------------
//	FUNCTIONS
//-----------------------------------------------------------------------------------

//-----------------------------------------------------------------------------------
//	SERVER DATE&TIME
//-----------------------------------------------------------------------------------
//	Get server time in string form

function get_server_time()
{
	my_date = new Date();

	return my_date.toUTCString();
}

//-----------------------------------------------------------------------------------
//	TIMESTAMP
//-----------------------------------------------------------------------------------
//	Profile performance in ms

function get_timestamp_ms()
{
	my_date = new Date();
	return 1000.0* my_date.getSeconds() +my_date.getMilliseconds()
}

//-----------------------------------------------------------------------------------
//	FILE LOADER
//-----------------------------------------------------------------------------------
//	Load files into memory for improved latency

function load_file( file_name )
{
	var file_tmp;
	var file_path =  __dirname +"/" +file_name;

	//HTML index file
	try
	{
		file_tmp = fs.readFileSync( file_path );
	}
	catch (err)
	{
		console.log("ERR: " +err.code +" failed to load: " +file_path);
		throw err;
	}

	console.log("INFO: " +file_path +" has been loaded into memory");

	return file_tmp;
}

//-----------------------------------------------------------------------------------
//	CONTENT TYPE DETECTOR
//-----------------------------------------------------------------------------------
//	Return the right content type to give correct information to the client browser

function detect_content( file_name )
{
	if (file_name.includes(".html"))
	{
        return "text/html";
	}
	else if (file_name.includes(".css"))
	{
		return "text/css";
	}
	else if (file_name.includes(".js"))
	{
		return "application/javascript";
	}
	else
	{
		throw "invalid extension";

	}
}

客户端 html

<!DOCTYPE html>
<meta charset="utf-8"/>
<html>
	<head>
		<title>Maze Runner</title>
		<link rel="stylesheet" href="style.css">
		<script type="text/javascript" src="/socket.io/socket.io.js"></script>
		<script type="text/javascript">
			var host_ip = document.location.hostname;
			console.log("connecting to host: ", host_ip);

			//Get references to the html controls
			textbox_input1 = window.document.getElementById("my_text_box")

			//Connect to the server via websocket
			var mysocket = io("http://" +host_ip +":8080");
			//Long lived frame object
			var last_frame;

			//-----------------------------------------
			//	CONNESSION ACKNOWLEDGE
			//-----------------------------------------
			//	Link is initiated by the client
			//	Server sends a welcome message when link is estabilished
			//	Server could send an auth token to keep track of individual clients and login data

			mysocket.on
			(
				"welcome",
				(message) =>
				{
					console.log("Server websocket connession acknoweldged... " +message.payload);
				}
			)

			//-----------------------------------------
			//	SERVER->CLIENT CONTROLS
			//-----------------------------------------
			//	Server can send an async message to dinamically update the page without reloading
			//	This is an example message with the server local date and time in string form

			mysocket.on
			(
				"server_time",
				(message) =>
				{
					fill_label( message.server_time );
					console.log("Server sent his local time... " +message.server_time);
				}
			)

			function fill_label( payload )
			{
				textbox_input1.value=payload;
			}

			//-----------------------------------------
			//	CLIENT->SERVER CONTROLS
			//-----------------------------------------
			//	Controls inside the webpage can emit async events to the server
			//	In this example I have a push button and I catch keyboard strokes

			//Handler for a pushbutton
			function socket_button_handler()
			{
				mysocket.emit("myclick", { payload: "button was clicked" });
				console.log("Button was clicked...");
			}

			//Listen for keystrokes
			window.document.addEventListener
			(
				"keypress",
				function onEvent(event)
				{
					//Inform the server that a key has been pressed
					mysocket.emit("keyboard", { payload: event.key });
					console.log("Key press...");
				}
			);

			//-----------------------------------------
			//	PING-PONG
			//-----------------------------------------
			//	Server sends ping messages with a timestamp
			//	Client answers with pongs to allow server to profile latency of the channel

			//profile messages means the server wants to compute roundway trip
			mysocket.on
			(
				"profile_ping",
				(message) =>
				{
					//Answer back with the received timestamp so that server can compute roundway trip
					mysocket.emit("profile_pong", { timestamp: message.timestamp });
					console.log( "server wants a pong. server absolute timestamp[ms]: " +message.timestamp );
				}
			);

		</script>
	</head>
	<body>

		<h1>Html+Css Server +low latency Websocket server</h1>
		<!-- button control with socket emitter as handler -->
		<p> This button will emit a websocket event. The server will be informed in real time of the event. </p>
		<button id="my_button" type="button" onclick="socket_button_handler()">Websocket Button!</button>

		<!-- input text control -->
		<p> This input can be filled through websockets directly by the server in real time </p>
		<input id="my_text_box" type="text" value="" size="40">

		<!-- canvas object, it's painted by the javascript video decoder -->
		<p> This canvas is painted by the javascript player and shows the live stream.'</p>
		<canvas id="video-canvas" width=640 height=480></canvas>

		<!-- Javascript video decoder, take in a data stream from a websocket and paint on a canvas -->
		<script type="text/javascript" src="jsmpeg.min.js"></script>
		<script type="text/javascript">
		var mycanvas = document.getElementById("video-canvas");
		var url = "ws://" + host_ip +":8082/";
		var player = new JSMpeg.Player(url, {canvas: mycanvas});
		</script>
	</body>
</html>

Javascript 播放器

你可以从这里获取我使用的javascript播放器： https://github.com/phoboslab/jsmpeg/blob/master/jsmpeg.min.js

Answer 2

我想要 NODE.JS 包或其他解决方案的建议，以提供 UDP H264 视频流，该视频流可以由 HTML5 视频标签解码，目标延迟为 50 毫秒。

在这种配置下这几乎肯定是不可能的。

如果您放弃视频标签要求，并在浏览器中直接使用 WebRTC，您可能可以将时间降至约 150 毫秒。