feat: rewrite project in C++ [closes #7]

.gitignore

@@ -1,3 +1,12 @@
+# C/C++ build artifacts
+build/
+cmake-build-*/
+CMakeCache.txt
+CMakeFiles/
+*.o
+*.a
+*.obj
+
 # Byte-compiled / optimized / DLL files
 __pycache__/
 *.py[cod]

CMakeLists.txt

@@ -0,0 +1,53 @@
+cmake_minimum_required(VERSION 3.16)
+project(videoanalytics LANGUAGES CXX)
+
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_CXX_STANDARD_REQUIRED ON)
+set(CMAKE_CXX_EXTENSIONS OFF)
+
+if(NOT CMAKE_BUILD_TYPE)
+    set(CMAKE_BUILD_TYPE Release)
+endif()
+
+find_package(OpenCV REQUIRED)
+find_package(Threads REQUIRED)
+find_package(nlohmann_json QUIET)
+
+# ZeroMQ (libzmq). Prefer pkg-config, fall back to plain library search.
+find_package(PkgConfig QUIET)
+if(PkgConfig_FOUND)
+    pkg_check_modules(ZMQ QUIET libzmq)
+endif()
+if(NOT ZMQ_FOUND)
+    find_path(ZMQ_INCLUDE_DIRS zmq.h)
+    find_library(ZMQ_LIBRARIES NAMES zmq libzmq)
+endif()
+if(NOT ZMQ_LIBRARIES)
+    message(FATAL_ERROR "libzmq not found. Install libzmq3-dev (or equivalent).")
+endif()
+
+add_library(va_core
+    src/config.cpp
+    src/camera.cpp
+    src/broker.cpp
+    src/worker.cpp
+    src/gui.cpp
+    src/psnr.cpp
+    src/detector.cpp
+)
+target_include_directories(va_core PUBLIC
+    ${CMAKE_CURRENT_SOURCE_DIR}/src
+    ${OpenCV_INCLUDE_DIRS}
+    ${ZMQ_INCLUDE_DIRS}
+)
+target_link_libraries(va_core PUBLIC
+    ${OpenCV_LIBS}
+    ${ZMQ_LIBRARIES}
+    Threads::Threads
+)
+if(nlohmann_json_FOUND)
+    target_link_libraries(va_core PUBLIC nlohmann_json::nlohmann_json)
+endif()
+
+add_executable(videoanalytics src/main.cpp)
+target_link_libraries(videoanalytics PRIVATE va_core)

README.md

@@ -1,52 +1,117 @@
 # videoanalytics
 
-This project is about videosources emergency detection made with ZeroMQ infrastructure
+This project is about videosources emergency detection made with ZeroMQ infrastructure.
+
+This is the C++ port of the project (originally written in Python). The
+architecture is unchanged: `cameras -> broker -> workers -> gui`, all connected
+over ZeroMQ.
 
 ## Used technologies
-- ZeroMQ
-- YOLOv8
+- ZeroMQ (libzmq / cppzmq)
+- YOLOv8 (via OpenCV DNN, ONNX models)
 - PSNR
 - OpenCV
-- Multiprocessing & Threading
+- C++17 threads
+
+## Architecture
+
+```
+cameras (PUB) --> broker (SUB -> PUSH) --> workers (PULL -> PUSH) --> gui (PULL)
+```
+
+- **cameras** – one thread per source reads a video and publishes JPEG frames
+  tagged with a 2-byte source id.
+- **broker** – subscribes to every source and routes each frame to the worker
+  whose index matches the source id.
+- **workers** – decode the frame, run the fire and forklift detectors plus PSNR
+  concurrently, and forward `[address, jpeg, detections_json, psnr]`.
+- **gui** – renders the latest annotated frame of every source on a grid with
+  bounding boxes and `WARNING` / `FIRE` overlays.
 
 ## Prerequisites
 
-- Torch with CUDA
-- Python3.12
+- A C++17 compiler
+- CMake >= 3.16
+- OpenCV 4 (with the `dnn` module)
+- ZeroMQ (`libzmq`) and the `cppzmq` header (`zmq.hpp`)
+- nlohmann/json
+
+On Debian/Ubuntu:
+
+```bash
+sudo apt-get install -y build-essential cmake libzmq3-dev libopencv-dev nlohmann-json3-dev pkg-config
+```
+
+## Models
+
+The original project shipped Ultralytics PyTorch checkpoints (`.pt`). The C++
+port runs the exported ONNX versions through OpenCV's DNN module. Export them
+once with Ultralytics:
+
+```bash
+yolo export model=forklift_8s.pt format=onnx
+yolo export model=fire.pt format=onnx
+```
+
+Place the resulting `.onnx` files under `./models` and point `config.json` at
+them. Class names default to `forklift` and `fire`; adjust them in
+`src/worker.cpp` if your models use different labels.
 
 ## Preparations
 
-- `pip install -r requirements.txt`
-- Download all necessary files from [Google Drive](https://drive.google.com/drive/u/0/folders/1OI_XtRNcwbm-JvojeKGR_x1SE1GuonQq) :
- - (optional) place test videos & meanframes to `./videos` folder
- - place YOLO models to `./models` folder
-- set up `config.json` with actual files locations, ip addresses & ports, psnr threshold
+- Download the necessary media/models and place test videos & meanframes in
+  `./videos` and YOLO ONNX models in `./models`.
+- Set up `config.json` with actual file locations, ip addresses & ports, and the
+  psnr threshold.
 
 Notes:
-> meanframes must have same resolution as original videos, '.jpg' hardcoded <BR>
-> meanframes were created on "normal" parts of videos with `psnr.py`
+> meanframes must have the same resolution as the original videos, '.jpg' hardcoded <BR>
+> meanframes are created on "normal" parts of videos with the `psnr` subcommand
 
-## Meanframes for PSNR on custom videos
-Create meanframes for PSNR function:
-- Prepare "normal" videofragment (not containing emergency situations) in video editor
-- Use `psnr.py`:
-    - `python psnr.py <videopath>`
-- Meanframe will be created in same directory with same filename in '.jpg' format
+## Build
+
+```bash
+mkdir -p build && cd build
+cmake ..
+make -j
+```
+
+This produces a single `videoanalytics` executable.
 
 ## Usage
-After you've set up config.json, just:
+
+Run the full pipeline (broker + workers + cameras + gui):
+
 ```bash
-python run.py
+./build/videoanalytics
 ```
-`Ctrl+C` to stop
+
+`ESC` (in the GUI window) to stop.
+
+Individual components can also be launched separately:
+
+```bash
+./build/videoanalytics broker
+./build/videoanalytics workers
+./build/videoanalytics cameras
+./build/videoanalytics gui
+```
+
+## Meanframes for PSNR on custom videos
+Create meanframes for the PSNR function:
+- Prepare a "normal" video fragment (not containing emergency situations).
+- Run the meanframe tool:
+    - `./build/videoanalytics psnr <videopath>`
+- The meanframe is created in the same directory with the same filename in
+  '.jpg' format.
 
 ## TODOs
 - create meanframes automatically
 - train models on another datasets
 - add resolution to config
 - Ctrl+C handler
-- Auto scale processes
+- Auto scale processes/threads
 - Test on realtime videosources
 
 ## Example
-![example](./media/example_videoanalytics.gif)
+![example](./media/example_videoanalytics.gif)

config.json

@@ -28,8 +28,8 @@
         "tcp://127.0.0.1:5565"
     ],
     "models": {
-        "fire": "./models/fire.pt",
-        "forklift": "./models/forklift_8s.pt"
+        "fire": "./models/fire.onnx",
+        "forklift": "./models/forklift_8s.onnx"
     },
     "gui": "tcp://127.0.0.1:5562",
     "psnr_threshold": 25