Vision Drive Assist — OpenGL HUD

Receives TCP JSON from the Python pipeline, renders a real-time ADAS HUD.

Dependencies

# Ubuntu / Debian
sudo apt install libglew-dev libglfw3-dev nlohmann-json3-dev

# Arch
sudo pacman -S glew glfw-x11 nlohmann-json

Build

mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
make -j$(nproc)

Run

# Start Python pipeline first (it sends TCP on port 5005)
python3 run/main.py --video my_drive.mp4

# In another terminal:
./build/adas_hud --port 5005

Keys

Key	Action
1	Toggle DA area
2	Toggle lane lines
3	Toggle cars
4	Toggle signs
D	Debug grid
S	Screenshot (.ppm)
Q / Esc	Quit

Integration with your existing code

tcp_client.cpp

Add your existing tcp_client.cpp to CMakeLists.txt under add_executable.

Box coordinate space

hud_renderer.cpp → drawCars() currently normalizes boxes assuming a 1920×1080 source frame:

float x1 = carBoxX(car.x1 / 1920.f);

If your frame is a different resolution, change those divisors, or normalize in parseObjects().

Lane coordinate space

Lane lines from HybridNets are in [0..1]×[0..1] within the lower 58% of the frame. laneToScreenY() maps ny=0 → 42% of screen height, ny=1 → bottom edge. Adjust roadTop if your pipeline uses a different crop.

Extending the template

Add text rendering (FreeType)

// In hud_renderer.h
#include <ft2build.h>
#include FT_FREETYPE_H
// Add TextRenderer member, call drawText("38 mph", x, y, size) from drawHUD()

Add ego car motion blur

// In drawEgoCar(): store last N positions, draw fading quads

Add distance estimation

// In drawCars(): estimate distance from box height
// dist_m ≈ (real_car_height_m * focal_length) / box_height_px