Compare commits

...

13 Commits

15 changed files with 791 additions and 192 deletions

3
.gitignore vendored
View File

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

View File

@ -3,38 +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
```shell
python -m client
```
### Server
```shell
python -m server
./build/client # for client application
./build/server # for server application
```

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;
}

View File

@ -1,76 +0,0 @@
import cv2
import socket
import numpy as np
import uuid
uuid = uuid.uuid4()
from common import StdPacket, InterlacedPacket, DoublyInterlacedPacket
UDP_IP = "127.0.0.1"
UDP_PORT = 5005
# Create a VideoCapture object
cap = cv2.VideoCapture(0) # 0 represents the default camera
# Check if camera opened successfully
if not cap.isOpened():
print("Error opening video stream or file")
def send_packet(sock, packet):
sock.sendto(packet, (UDP_IP, UDP_PORT))
sock = socket.socket(socket.AF_INET, # Internet
socket.SOCK_DGRAM) # UDP
def breakdown_image_norm(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])
send_packet(sock, pkt.to_bytestr())
def breakdown_image_interlaced(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])
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])
send_packet(sock, pkt.to_bytestr())
def breakdown_image_dint(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 l in range(0, 4):
for i in range(0, cols, 32):
for j in range(0, rows, 32):
# 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])
send_packet(sock, pkt.to_bytestr())
while True:
# Capture frame-by-frame
ret, frame = cap.read()
# If frame is read correctly, ret is True
if not ret:
print("Can't receive frame (stream end?). Exiting ...")
break
breakdown_image_dint(frame)
# Release the capture and close all windows
cap.release()

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` |

129
legacy/client.py Normal file
View File

@ -0,0 +1,129 @@
import argparse
import cv2
import socket
import numpy as np
import uuid
from common import StdPacket, InterlacedPacket, DoublyInterlacedPacket, TiledImagePacket
def send_packet(sock, packet):
sock.sendto(packet, (UDP_IP, UDP_PORT))
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)
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, 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)
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)
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, 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 l in range(0, 4):
for i in range(0, cols, 32):
for j in range(0, rows, 32):
# print("Sending frame segment (%d, %d)", i, j)
i_even = l % 2 == 0
j_even = l >= 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__':
# argument parser
parser = argparse.ArgumentParser(description="Proof-of-concept client for sauron-cv")
parser.add_argument("-p", "--port", type=int, default=5005)
parser.add_argument("-s", "--server", type=str, default="127.0.0.1")
parser.add_argument("-W", "--width", type=int, default=640)
parser.add_argument("-H", "--height", type=int, default=480)
parser.add_argument("-d", "--device", type=int, default=0)
args = parser.parse_args()
# give this client its very own UUID
uuid = uuid.uuid4()
# give target IP address
UDP_IP = args.server
UDP_PORT = args.port
WIDTH = args.width
HEIGHT = args.height
DEVICE = args.device
# Create a VideoCapture object
cap = cv2.VideoCapture(DEVICE) # 0 represents the default camera
cap.set(cv2.CAP_PROP_FRAME_WIDTH, WIDTH)
cap.set(cv2.CAP_PROP_FRAME_HEIGHT, HEIGHT)
# Check if camera opened successfully
if not cap.isOpened():
print("Error opening video stream or file")
# 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()
# If frame is read correctly, ret is True
if not ret:
print("Can't receive frame (stream end?). Exiting ...")
break
breakdown_image_dint(frame, last_frame)
# Release the capture and close all windows
cap.release()

View File

@ -3,7 +3,18 @@ from abc import ABC, abstractmethod
import numpy as np
from uuid import UUID
# The basic structure of the packet is as follows:
# FFFF FFFF FFFF FFFF FFFF FFFF FFFF FFFF FFFF FFFF FFFF FFFF FFFF ... FFFF
# | uuid | | | |
# | | | | image data: contains the 16x16 image slice bitpacked from a NumPy array
# | | | y: the y position of this packet in the original image
# | | x: the x position of this packet in the original image
# | uuid: matches the packet to the requesting client
# Other packet types may change this structure. Need to standardize this somehow.
class Packet(ABC):
"""Generic structure for a video streaming packet. Contains a slice of the full image, which will be reconstructed
by the server."""
size: int
def __init__(self, uuid: UUID, x: int, y: int, array: np.ndarray):
self.uuid = uuid
@ -13,13 +24,16 @@ class Packet(ABC):
@abstractmethod
def to_bytestr(self) -> bytes:
"""Convert a packet object into a bytestring."""
pass
@abstractmethod
def apply(self, image: np.ndarray) -> np.ndarray:
"""Apply this packet to an image."""
pass
class StdPacket(Packet):
"""A standard packet with no interlacing. Sends a 16x16 chunk of an image."""
size = 16 + 4 + 4 + 768
def __init__(self, uuid: UUID, x: int, y: int, array: np.ndarray):
super().__init__(uuid, x, y, array)
@ -40,6 +54,7 @@ class StdPacket(Packet):
return image
def from_bytes_std(b: bytes) -> StdPacket:
"""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)
@ -49,6 +64,7 @@ def from_bytes_std(b: bytes) -> StdPacket:
class InterlacedPacket(Packet):
"""A packet with horizontal interlacing. Sends half of a 16x32 chunk of an image, every other row"""
size = 16 + 4 + 4 + 4 + 768
def __init__(self, uuid: UUID, x: int, y: int, even: bool, array: np.ndarray):
super().__init__(uuid, x, y, array)
@ -72,6 +88,7 @@ class InterlacedPacket(Packet):
def from_bytes_int(b: bytes) -> InterlacedPacket:
"""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)
@ -81,6 +98,8 @@ def from_bytes_int(b: bytes) -> InterlacedPacket:
return InterlacedPacket(uuid, x, y, even, array)
class DoublyInterlacedPacket(Packet):
"""A packet with horizontal interlacing. Sends one quarter of a 32x32 chunk of an image. This will alternate rows
and columns based on the value of even_x and even_y."""
size = 16 + 4 + 4 + 4 + 768
def __init__(self, uuid: UUID, x: int, y: int, even_x: bool, even_y: bool, array: np.ndarray):
super().__init__(uuid, x, y, array)
@ -106,6 +125,7 @@ class DoublyInterlacedPacket(Packet):
def from_bytes_dint(b: bytes) -> DoublyInterlacedPacket:
"""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)
@ -113,4 +133,39 @@ def from_bytes_dint(b: bytes) -> DoublyInterlacedPacket:
even_y = bool.from_bytes(b[26:28])
array = np.frombuffer(b[28:], np.uint8).reshape(16, 16, 3)
return DoublyInterlacedPacket(uuid, x, y, even_x, even_y, array)
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)

4
legacy/requirements.txt Normal file
View File

@ -0,0 +1,4 @@
opencv-python
numpy
rich
asyncudp

112
legacy/server.py Normal file
View File

@ -0,0 +1,112 @@
from typing import Dict
import cv2
import socket
import numpy as np
from datetime import datetime
import asyncio
import asyncudp
import argparse
from rich.console import Console
from common import Packet, DoublyInterlacedPacket, from_bytes_dint
class Client:
"""Class for tracking client state, including current frame data and time since last update."""
def __init__(self):
self.last_updated = datetime.now()
self.frame = np.ndarray((HEIGHT, WIDTH, 3), dtype=np.uint8)
def update(self, pkt: Packet):
"""Apply a packet to the client frame. Update last client update to current time."""
self.frame = pkt.apply(self.frame)
self.last_updated = datetime.now()
def latency(self) -> float:
"""Return the time since the last client update."""
return (datetime.now() - self.last_updated).total_seconds()
def read(self) -> np.ndarray:
"""Return the current frame."""
return self.frame
# Dictionary of client states stored by UUID
frames: Dict[str, Client] = {}
async def show_frames():
"""Asynchronous coroutine to display frames in OpenCV debug windows."""
while True:
# drop clients that have not sent packets for > 5 seconds
for id in list(frames.keys()):
if frames[id].latency() >= 5:
console.log(f"Client likely lost connection, dropping [bold red]{id}[/bold red]")
cv2.destroyWindow(id)
frames.pop(id)
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(0.05)
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="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()
# console
console = Console()
# assign constants based on argument parser
UDP_IP = args.listen
UDP_PORT = args.port
HEIGHT = args.height
WIDTH = args.width
# 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(listen(UDP_IP, UDP_PORT))
loop.create_task(show_frames())
try:
loop.run_forever()
finally:
loop.run_until_complete(loop.shutdown_asyncgens())
loop.close()
# Release the capture and close all windows
cv2.destroyAllWindows()

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])

View File

@ -1,2 +0,0 @@
opencv-python
numpy

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;
}

View File

@ -1,91 +0,0 @@
from typing import Dict
import cv2
import socket
import numpy as np
from datetime import datetime
import asyncio
from common import DoublyInterlacedPacket, from_bytes_dint
# bind any IP address
UDP_IP = ""
UDP_PORT = 5005
HEIGHT = 480
WIDTH = 640
class Client:
def __init__(self):
self.last_updated = datetime.now()
self.frame = np.ndarray((HEIGHT, WIDTH, 3), dtype=np.uint8)
def update(self, pkt: DoublyInterlacedPacket):
self.frame = pkt.apply(self.frame)
self.last_updated = datetime.now()
def latency(self) -> float:
return (datetime.now() - self.last_updated).total_seconds()
def read(self) -> np.ndarray:
return self.frame
frames: Dict[str, Client] = {}
async def read_packet():
while True:
for i in range(0, 1200):
# break the array down into 16-bit chunks, then transmit them as UDP packets
try:
data, addr = sock.recvfrom(DoublyInterlacedPacket.size) # buffer size is 768 bytes
# print("received packet from", addr)
if data:
pkt = from_bytes_dint(data)
uuid = str(pkt.uuid)
if uuid not in frames.keys():
print("New client acquired, naming %s", uuid)
frames[uuid] = Client()
frames[uuid].update(pkt)
except BlockingIOError:
pass
await asyncio.sleep(0.001)
async def show_frames():
while True:
# Display the resulting frame
for id in list(frames.keys()):
if frames[id].latency() >= 5:
print("Client likely lost connection, dropping %s", id)
cv2.destroyWindow(id)
frames.pop(id)
else:
cv2.imshow(id, frames[id].read())
cv2.waitKey(1)
await asyncio.sleep(0.01)
if __name__ == "__main__":
sock = socket.socket(socket.AF_INET, # Internet
socket.SOCK_DGRAM) # UDP
sock.setblocking(False)
sock.bind((UDP_IP, UDP_PORT))
loop = asyncio.new_event_loop()
asyncio.set_event_loop(loop)
loop.create_task(read_packet())
loop.create_task(show_frames())
try:
loop.run_forever()
finally:
loop.run_until_complete(loop.shutdown_asyncgens())
loop.close()
# Release the capture and close all windows
cv2.destroyAllWindows()

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