Compare commits

...

11 Commits

12 changed files with 616 additions and 109 deletions

1
.gitignore vendored
View File

@ -1,2 +1,3 @@
.venv
.idea
build

View File

@ -3,56 +3,37 @@
This project is a demo for streaming video output from multiple "client" devices to one "server". This is a basic demo
of what sauron-cv seeks to accomplish.
## Installation
## Building
It is strongly recommended that you use a virtual environment. These instructions will assume you are using venv, you
can substitute this with your preferred environment management. You will need to make sure that the `virtualenv` package
is installed globally, either via pip or the `python3-virtualenv` package in your system package manager.
This project uses Meson for build management.
When first cloning this repo, run the following:
### Install Dependencies
Install the following packages from your distribution package manager:
- `meson`
- `opencv`
- `boost`
A better procedure for this (hopefully involving Meson) will be added / documented at a later date.
### Setup Meson
```shell
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
meson setup build
```
This will create a virtual environment, enter that virtual environment, and install the required packages.
*NOTE FOR JETBRAINS / CLION USERS: PLEASE SET YOUR MESON BUILD DIRECTORY TO `build` IN THE IDE SETTINGS UNDER "Build /
Execution / Deployment" -> "Meson"*
If you start a new shell, you will need to re-enter the virtual environment:
### Compiling
```shell
source .venv/bin/activate
meson build client # for client only
meson build server # for server only
```
## Running
### Client
To run the client with a localhost target:
```shell
python -m client
./build/client # for client application
./build/server # for server application
```
To target an external server, provide its IP address using the `-s` flag:
```shell
python -m client -s [server IP address]
```
### Server
```shell
python -m server
```
### Common Flags
Make sure that these match between your client and server!!
| Short | Long | Description | Default |
|-------|---|---|---|
| `-p` | `--port` | The port for the client / server to communicate on. | `5005` |
| `-W` | `--width` | Image width in pixels. | `640` |
| `-H` | `--height` | Image height in pixels. | `480` |

43
client.cpp Normal file
View File

@ -0,0 +1,43 @@
#include <opencv2/videoio.hpp>
#include <cstring>
#include <iostream>
#include <netinet/in.h>
#include <sys/socket.h>
#include <unistd.h>
#include "transfer.h"
using namespace std;
int main() {
// create video capture
cv::VideoCapture cap = cv::VideoCapture(0);
// create socket
int clientSocket = socket(AF_INET, SOCK_STREAM, 0);
// specifying address
sockaddr_in serverAddress;
serverAddress.sin_family = AF_INET;
serverAddress.sin_port = htons(8080);
serverAddress.sin_addr.s_addr = INADDR_ANY;
cv::Mat image = cv::Mat::zeros(cv::Size(640, 480), CV_8UC3);
info("Ready to connect.");
// sending connection request
connect(clientSocket, reinterpret_cast<sockaddr *>(&serverAddress),
sizeof(serverAddress));
info("Connected.");
// create buffer for serialization
vector<uchar> imgbuf;
while (true) {
cap.read(image);
trace("Sending image");
sendImage(clientSocket, image, imgbuf);
}
close(clientSocket);
return 0;
}

58
legacy/README.md Normal file
View File

@ -0,0 +1,58 @@
# Video Streaming Proof of Concept
This project is a demo for streaming video output from multiple "client" devices to one "server". This is a basic demo
of what sauron-cv seeks to accomplish.
## Installation
It is strongly recommended that you use a virtual environment. These instructions will assume you are using venv, you
can substitute this with your preferred environment management. You will need to make sure that the `virtualenv` package
is installed globally, either via pip or the `python3-virtualenv` package in your system package manager.
When first cloning this repo, run the following:
```shell
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
```
This will create a virtual environment, enter that virtual environment, and install the required packages.
If you start a new shell, you will need to re-enter the virtual environment:
```shell
source .venv/bin/activate
```
## Running
### Client
To run the client with a localhost target:
```shell
python -m client
```
To target an external server, provide its IP address using the `-s` flag:
```shell
python -m client -s [server IP address]
```
### Server
```shell
python -m server
```
### Common Flags
Make sure that these match between your client and server!!
| Short | Long | Description | Default |
|-------|---|---|---|
| `-p` | `--port` | The port for the client / server to communicate on. | `5005` |
| `-W` | `--width` | Image width in pixels. | `640` |
| `-H` | `--height` | Image height in pixels. | `480` |

View File

@ -5,37 +5,49 @@ import socket
import numpy as np
import uuid
from common import StdPacket, InterlacedPacket, DoublyInterlacedPacket
from common import StdPacket, InterlacedPacket, DoublyInterlacedPacket, TiledImagePacket
def send_packet(sock, packet):
sock.sendto(packet, (UDP_IP, UDP_PORT))
def breakdown_image_norm(frame):
def breakdown_image_norm(frame, last_frame):
(cols, rows, colors) = frame.shape
# break the array down into 16x16 chunks, then transmit them as UDP packets
for i in range(0, cols, 16):
for j in range(0, rows, 16):
# print("Sending frame segment (%d, %d)", i, j)
pkt = StdPacket(uuid, j, i, frame[i:i + 16, j:j + 16])
arr = frame[i:i + 16, j:j + 16]
last_arr = last_frame[i:i + 16, j:j + 16]
# only update if image segments are different
if not np.allclose(arr, last_arr):
pkt = StdPacket(uuid, j, i, arr)
send_packet(sock, pkt.to_bytestr())
def breakdown_image_interlaced(frame):
def breakdown_image_interlaced(frame, last_frame):
(cols, rows, colors) = frame.shape
# break the array into 16x32 chunks. we'll split those further into odd and even rows
# and send each as UDP packets. this should make packet loss less obvious
for i in range(0, cols, 32):
for j in range(0, rows, 16):
# print("Sending frame segment (%d, %d)", i, j)
pkt = InterlacedPacket(uuid, j, i, False, frame[i:i + 32:2, j:j + 16])
arr = frame[i:i + 32:2, j:j + 16]
last_arr = last_frame[i:i + 32:2, j:j + 16]
if not np.allclose(arr, last_arr):
pkt = InterlacedPacket(uuid, j, i, False, arr)
send_packet(sock, pkt.to_bytestr())
for i in range(0, cols, 32):
for j in range(0, rows, 16):
# print("Sending frame segment (%d, %d)", i, j)
pkt = InterlacedPacket(uuid, j, i, True, frame[i + 1:i + 32:2, j:j + 16])
arr = frame[i + 1:i + 32:2, j:j + 16]
last_arr = last_frame[i + 1:i + 32:2, j:j + 16]
# only update if image segments are different
if not np.allclose(arr, last_arr):
pkt = InterlacedPacket(uuid, j, i, True, arr)
send_packet(sock, pkt.to_bytestr())
def breakdown_image_dint(frame):
def breakdown_image_dint(frame, last_frame):
(cols, rows, colors) = frame.shape
# break the array into 16x32 chunks. we'll split those further into odd and even rows
# and send each as UDP packets. this should make packet loss less obvious
@ -45,7 +57,25 @@ def breakdown_image_dint(frame):
# print("Sending frame segment (%d, %d)", i, j)
i_even = l % 2 == 0
j_even = l >= 2
pkt = DoublyInterlacedPacket(uuid, j, i, j_even, i_even, frame[i + i_even:i + 32:2, j + j_even:j + 32:2])
# breakdown image
arr = frame[i + i_even:i + 32:2, j + j_even:j + 32:2]
last_arr = last_frame[i + i_even:i + 32:2, j + j_even:j + 32:2]
# only update if image segments are different
if not np.allclose(arr, last_arr):
pkt = DoublyInterlacedPacket(uuid, j, i, j_even, i_even, arr)
send_packet(sock, pkt.to_bytestr())
def breakdown_image_tiled(frame):
(cols, rows, colors) = frame.shape
# break the array into 16x32 chunks. we'll split those further into odd and even rows
# and send each as UDP packets. this should make packet loss less obvious
xslice = cols // 16
yslice = rows // 16
for i in range(0, xslice):
for j in range(0, yslice):
# print("Sending frame segment (%d, %d)", i, j)
pkt = TiledImagePacket(uuid, j, i, rows, cols, frame[i:cols:xslice, j:rows:yslice])
send_packet(sock, pkt.to_bytestr())
if __name__ == '__main__':
@ -80,7 +110,11 @@ if __name__ == '__main__':
# create the socket
sock = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
frame = np.zeros((HEIGHT, WIDTH, 3), dtype=np.uint8)
last_frame = np.zeros((HEIGHT, WIDTH, 3), dtype=np.uint8)
while True:
last_frame = frame.copy()
# Capture frame-by-frame
ret, frame = cap.read()
@ -89,7 +123,7 @@ if __name__ == '__main__':
print("Can't receive frame (stream end?). Exiting ...")
break
breakdown_image_dint(frame)
breakdown_image_dint(frame, last_frame)
# Release the capture and close all windows
cap.release()

View File

@ -134,3 +134,38 @@ def from_bytes_dint(b: bytes) -> DoublyInterlacedPacket:
array = np.frombuffer(b[28:], np.uint8).reshape(16, 16, 3)
return DoublyInterlacedPacket(uuid, x, y, even_x, even_y, array)
class TiledImagePacket(Packet):
"""Distributed selection from image."""
size = 16 + 4 + 4 + 768
def __init__(self, uuid: UUID, x: int, y: int, width: int, height: int, array: np.ndarray):
super().__init__(uuid, x, y, array)
self.width = width
self.height = height
self.xslice = width // 16
self.yslice = height // 16
def to_bytestr(self) -> bytes:
bytestr = b""
bytestr += self.uuid.bytes
bytestr += self.x.to_bytes(length=4, signed=False)
bytestr += self.y.to_bytes(length=4, signed=False)
bytestr += self.array.tobytes()
return bytestr
def apply(self, image: np.ndarray) -> np.ndarray:
x = self.x
y = self.y
arr = self.array
image[y:self.height:self.yslice, x:self.width:self.xslice] = arr
return image
def from_bytes_tiled(b: bytes) -> TiledImagePacket:
"""Convert a byte string obtained via UDP into a packet object."""
uuid = UUID(bytes = b[0:16])
x = int.from_bytes(b[16:20], signed = False)
y = int.from_bytes(b[20:24], signed = False)
array = np.frombuffer(b[24:], np.uint8).reshape(16, 16, 3)
return TiledImagePacket(uuid, x, y, 640, 480, array)

View File

@ -1,3 +1,4 @@
opencv-python
numpy
rich
asyncudp

View File

@ -5,6 +5,7 @@ import socket
import numpy as np
from datetime import datetime
import asyncio
import asyncudp
import argparse
from rich.console import Console
@ -34,33 +35,6 @@ class Client:
# Dictionary of client states stored by UUID
frames: Dict[str, Client] = {}
async def read_packet():
"""Asynchronous coroutine to read UDP packets from the client(s)."""
while True:
# we repeat this a ton of times all at once to hopefully capture all of the image data
for i in range(0, 1600):
try:
data, addr = sock.recvfrom(DoublyInterlacedPacket.size) # packet buffer size based on the packet size
# print("received packet from", addr)
if data:
pkt = from_bytes_dint(data)
uuid = str(pkt.uuid)
# if this is a new client, give it a new image
if uuid not in frames.keys():
console.log(f"New client acquired, naming [bold cyan]{uuid}[bold cyan]")
frames[uuid] = Client()
stat.update(f"[bold yellow]{len(frames.keys())}[/bold yellow] clients connected.")
frames[uuid].update(pkt)
except BlockingIOError:
pass
# this is necessary to allow asyncio to swap between reading packets and rendering frames
await asyncio.sleep(0.001)
async def show_frames():
"""Asynchronous coroutine to display frames in OpenCV debug windows."""
while True:
@ -70,20 +44,43 @@ async def show_frames():
console.log(f"Client likely lost connection, dropping [bold red]{id}[/bold red]")
cv2.destroyWindow(id)
frames.pop(id)
stat.update(f"[bold yellow]{len(frames.keys())}[/bold yellow] clients connected.")
else:
# show the latest available frame
cv2.imshow(id, frames[id].read())
cv2.waitKey(1)
# this is necessary to allow asyncio to swap between reading packets and rendering frames
await asyncio.sleep(0.01)
await asyncio.sleep(0.05)
async def listen(ip: str, port: int):
"""Asynchronous coroutine to listen for / read client connections."""
sock = await asyncudp.create_socket(local_addr=(ip, port))
console.log("Ready to accept connections.", style="bold green")
while True:
# receive packets
data, addr = await sock.recvfrom()
if data:
# convert the byte string into a packet object
pkt = from_bytes_dint(data)
uuid = str(pkt.uuid)
# if this is a new client, give it a new image
if uuid not in frames.keys():
console.log(f"New client acquired, naming [bold cyan]{uuid}[bold cyan]")
frames[uuid] = Client()
frames[uuid].update(pkt)
if __name__ == "__main__":
# argument parser
parser = argparse.ArgumentParser(description="Proof-of-concept server for sauron-cv")
parser.add_argument("-p", "--port", type=int, default=5005)
parser.add_argument("-l", "--listen", type=str, default="")
parser.add_argument("-l", "--listen", type=str, default="0.0.0.0")
parser.add_argument("-W", "--width", type=int, default=640)
parser.add_argument("-H", "--height", type=int, default=480)
args = parser.parse_args()
@ -98,22 +95,12 @@ if __name__ == "__main__":
HEIGHT = args.height
WIDTH = args.width
# create the UDP socket
sock = socket.socket(socket.AF_INET, # Internet
socket.SOCK_DGRAM) # UDP
sock.setblocking(False)
sock.bind((UDP_IP, UDP_PORT))
console.log("Ready to accept connections.", style="bold green")
with console.status("[bold yellow]0[/bold yellow] clients connected.", spinner="pong") as stat:
# create the async event loop
loop = asyncio.new_event_loop()
asyncio.set_event_loop(loop)
# create async tasks for reading network packets, displaying windows
loop.create_task(read_packet())
loop.create_task(listen(UDP_IP, UDP_PORT))
loop.create_task(show_frames())
try:
loop.run_forever()

178
logging.h Normal file
View File

@ -0,0 +1,178 @@
#ifndef LOGGING_H
#define LOGGING_H
#include <iostream>
#include <iomanip>
#include <chrono>
#include <ctime>
// Logging levels
#define LOG_ERROR 0
#define LOG_WARN 1
#define LOG_INFO 2
#define LOG_DEBUG 3
#define LOG_TRACE 4
// Set logging arguments
// TODO: set these by compiler arguments or env vars or something
#define LOGGING LOG_DEBUG // logging level
#define LOGGING_COLOR true // enable color
#define LOGGING_TIMESTAMP true // enable timestamp
#define LOGGING_TIMESTAMP_FMT "%Y-%m-%dT%H:%M:%S%z" // timestamp format (local time)
#define LOGGING_POSITION true // display position (only works on C++20 or newer)
// Color codes
#define ANSI_RESET "\033[0m"
#define ANSI_BLACK "\033[30m" /* Black */
#define ANSI_RED "\033[31m" /* Red */
#define ANSI_GREEN "\033[32m" /* Green */
#define ANSI_YELLOW "\033[33m" /* Yellow */
#define ANSI_BLUE "\033[34m" /* Blue */
#define ANSI_MAGENTA "\033[35m" /* Magenta */
#define ANSI_CYAN "\033[36m" /* Cyan */
#define ANSI_WHITE "\033[37m" /* White */
#define ANSI_BOLD "\033[1m" /* Bold */
template <typename T>
void print(T t) {
std::cout << t << std::endl;
}
template <typename T, typename... Args>
void print(T t, Args... args) {
std::cout << t;
print(args...);
}
inline void printTimestamp() {
#if LOGGING_TIMESTAMP
auto now = std::chrono::system_clock::now();
auto time_c = std::chrono::system_clock::to_time_t(now);
std::tm time_tm;
localtime_r(&time_c, &time_tm);
std::cout << std::put_time(&time_tm, LOGGING_TIMESTAMP_FMT) << ": ";
#endif
}
// if we're on C++20 or later, then use the source_location header and add source location to logs
#if __cplusplus >= 202002L
#include <source_location>
inline void printPosition(std::source_location& location) {
#if LOGGING_POSITION
std::cout << location.file_name() << ":" << location.function_name << ":" << ANSI_CYAN << location.line() << ANSI_RESET << ": ";
#endif
}
inline void printHeader(std::string name, std::string color, std::source_location& location) {
#if LOGGING_COLOR
std::cout << ANSI_BOLD << color << "[" << name << "] " << ANSI_RESET;
printTimestamp();
printPosition(location);
#else
printHeader(name);
#endif
}
inline void printHeader(std::string name, std::source_location& location) {
std::cout << "[" << name << "] ";
printTimestamp();
printPosition(location);
}
template <typename... Args, typename Sl = std::source_location>
void trace(Args... args, Sl location = std::source_location::current()) {
#if LOGGING >= LOG_TRACE
printHeader("TRACE", ANSI_CYAN, location);
print(args...);
#endif
}
template <typename... Args, typename Sl = std::source_location>
void debug(Args... args, Sl location = std::source_location::current()) {
#if LOGGING >= LOG_DEBUG
printHeader("DEBUG", ANSI_MAGENTA, location);
print(args...);
#endif
}
template <typename... Args, typename Sl = std::source_location>
void info(Args... args, Sl location = std::source_location::current()) {
#if LOGGING >= LOG_INFO
printHeader("INFO", ANSI_GREEN, location);
print(args...);
#endif
}
template <typename... Args, typename Sl = std::source_location>
void warn(Args... args, Sl location = std::source_location::current()) {
#if LOGGING >= LOG_WARN
printHeader("WARN", ANSI_YELLOW, location);
print(args...);
#endif
}
template <typename... Args, typename Sl = std::source_location>
void error(Args... args, Sl location = std::source_location::current()) {
#if LOGGING >= LOG_ERROR
printHeader("ERROR", ANSI_RED, location);
print(args...);
#endif
}
#else
inline void printHeader(std::string name, std::string color) {
#if LOGGING_COLOR
std::cout << ANSI_BOLD << color << "[" << name << "] " << ANSI_RESET;
printTimestamp();
#else
printHeader(name);
#endif
}
inline void printHeader(std::string name) {
std::cout << "[" << name << "] ";
printTimestamp();
}
template <typename... Args>
void trace(Args... args) {
#if LOGGING >= LOG_TRACE
printHeader("TRACE", ANSI_CYAN);
print(args...);
#endif
}
template <typename... Args>
void debug(Args... args) {
#if LOGGING >= LOG_DEBUG
printHeader("DEBUG", ANSI_MAGENTA);
print(args...);
#endif
}
template <typename... Args>
void info(Args... args) {
#if LOGGING >= LOG_INFO
printHeader("INFO", ANSI_GREEN);
print(args...);
#endif
}
template <typename... Args>
void warn(Args... args) {
#if LOGGING >= LOG_WARN
printHeader("WARN", ANSI_YELLOW);
print(args...);
#endif
}
template <typename... Args>
void error(Args... args) {
#if LOGGING >= LOG_ERROR
printHeader("ERROR", ANSI_RED);
print(args...);
#endif
}
#endif
#endif //LOGGING_H

33
meson.build Normal file
View File

@ -0,0 +1,33 @@
#=======================================================================================================================
# PROJECT SETTINGS
#=======================================================================================================================
project('video-streaming-poc', 'cpp', version : '0.0.1-SNAPSHOT',
default_options : ['c_std=c17', 'cpp_std=c++20'])
#=======================================================================================================================
# DEPENDENCIES
#=======================================================================================================================
# opencv dependency
opencv = dependency('opencv4', version : '>=4.0.0')
# boost dependency
boost = dependency('boost')
#=======================================================================================================================
# SOURCE FILES
#=======================================================================================================================
# common files between client / server
common = ['transfer.h', 'logging.h']
# client-only files
client = common + ['client.cpp']
# server-only files
server = common + ['server.cpp']
#=======================================================================================================================
# BUILD TARGETS
#=======================================================================================================================
# client executable
client_exe = executable('client', client,
dependencies : [opencv, boost])
# server executable
server_exe = executable('server', server,
dependencies : [opencv, boost])

55
server.cpp Normal file
View File

@ -0,0 +1,55 @@
#include <opencv2/highgui.hpp>
#include <opencv2/core/mat.hpp>
#include <cstring>
#include <iostream>
#include <netinet/in.h>
#include <sys/socket.h>
#include <unistd.h>
#include "transfer.h"
using namespace std;
int main() {
// TODO: read image data from socket instead of VideoCapture
// creating socket
int serverSocket = socket(AF_INET, SOCK_STREAM, 0);
// specifying the address
sockaddr_in serverAddress;
serverAddress.sin_family = AF_INET;
serverAddress.sin_port = htons(8080);
serverAddress.sin_addr.s_addr = INADDR_ANY;
// binding socket.
bind(serverSocket, reinterpret_cast<sockaddr *>(&serverAddress),
sizeof(serverAddress));
info("Ready to accept connections.");
// listening to the assigned socket
listen(serverSocket, 5);
// accepting connection request
int clientSocket = accept(serverSocket, nullptr, nullptr);
info("Client connected.");
// TODO: handle multiple images
cv::Mat image = cv::Mat::zeros(cv::Size(640, 480), CV_8UC3);
bool running = true;
// TODO: make this asynchronous. probably do that in tandem with setting up networking
while (running) {
// receive data
vector<uchar> buffer;
trace("Receiving image");
recvImage(clientSocket, buffer);
trace("Applying new data to image");
applyImage(image, &buffer);
trace("Displaying image");
imshow("image", image);
running = cv::waitKey(30) != 27;
}
close(serverSocket);
return 0;
}

101
transfer.h Normal file
View File

@ -0,0 +1,101 @@
#ifndef TRANSFER_H
#define TRANSFER_H
#include <chrono>
#include <opencv2/core/mat.hpp>
#include <opencv2/imgcodecs.hpp>
#include <sys/socket.h>
#include <logging.h>
#include <vector>
using namespace std;
struct imageHeader {
size_t size;
chrono::milliseconds timestamp;
};
chrono::milliseconds getMillis() {
// Get current time
const auto now = chrono::system_clock::now();
// Get time since epoch in milliseconds
const chrono::duration ms = chrono::duration_cast<chrono::milliseconds>(now.time_since_epoch());
return chrono::milliseconds(ms.count());
}
inline void serializeImage(const cv::Mat& image, std::vector<uchar>& buffer) {
cv::imencode(".jpg", image, buffer);
}
int sendImage(int socket, const cv::Mat& image, std::vector<uchar>& buffer) {
serializeImage(image, buffer);
size_t totalSent = 0;
// first send the size of the serialized image
const size_t size = buffer.size();
const chrono::milliseconds timestamp = getMillis();
imageHeader header;
header.size = size;
header.timestamp = timestamp;
trace("Buffer size: ", size);
if (const ssize_t sent = send(socket, &header, sizeof(header), 0); sent == -1) {
error("Error sending data header");
return -1;
}
// then start sending the serialized image
while (totalSent < size) {
const ssize_t sent = send(socket, buffer.data() + totalSent, size - totalSent, 0);
if (sent == -1) {
error("Error sending data");
return -1;
}
totalSent += sent;
debug("Packet sent (", sent, " bytes, total ", totalSent, " bytes)");
}
return 0;
}
int recvImage(int socket, std::vector<uchar>& buffer) {
// first receive the size of the image
imageHeader header;
if (ssize_t recvd = recv(socket, &header, sizeof(imageHeader), 0); recvd <= 0) {
error("Error receiving data header");
return -1;
}
size_t dataSize = header.size;
chrono::milliseconds sentTime = header.timestamp;
trace("Buffer size: ", dataSize);
// resize the buffer to fit the whole image
buffer.resize(dataSize);
// start receiving the image until the whole thing arrives
size_t totalReceived = 0;
while (totalReceived < dataSize) {
ssize_t bytesReceived = recv(socket, buffer.data() + totalReceived, dataSize, 0);
if (bytesReceived <= 0) {
error("Error receiving data");
buffer.clear();
return -1;
}
totalReceived += bytesReceived;
debug("Packet received (", bytesReceived, " bytes, total ", totalReceived, " bytes)");
}
chrono::milliseconds currentTime = getMillis();
chrono::milliseconds diff = currentTime - sentTime;
debug("Packet latency: ", diff.count(), "ms");
return 0;
}
bool applyImage(cv::Mat& image, std::vector<uchar> *src) {
// decode the image into an OpenCV Mat
cv::imdecode(*src, cv::IMREAD_UNCHANGED, &image);
return true;
}
#endif //TRANSFER_H