trajviz

Fast 3D camera trajectory video renderer using NumPy + OpenCV.

Renders animated top-down trajectory visualizations from 3D camera positions, with configurable ghost trails, color-coded markers, glow effects, and camera frustum overlays. Frames are piped directly to ffmpeg as raw RGB — no intermediate files.

Spiral trajectory with wedge frustum overlay. Colors encode time progression: blue at the start, magenta at the end. Generated by scripts/generate_demo.py.

Colors are user-provided as RGBA per position (n, 4), so they can encode anything. The renderer draws each segment with its assigned color and smoothly advances through the trajectory.

Coloring strategies

Time gradient — show trajectory progression:

t = np.linspace(0, 1, n, dtype=np.float32)
colors = np.stack([t, np.zeros(n), 1 - t, np.ones(n)], axis=1, dtype=np.float32)

Floor / zone coloring — assign colors by vertical position:

colors = np.zeros((n, 4), dtype=np.float32)
colors[:, 3] = 1.0  # full opacity
for i, z in enumerate(positions[:, 2]):
    if z < 3.0:
        colors[i, :3] = [0.2, 0.5, 0.9]   # blue = ground floor
    elif z < 6.0:
        colors[i, :3] = [0.9, 0.5, 0.1]   # orange = first floor
    else:
        colors[i, :3] = [0.1, 0.8, 0.4]   # green = second floor

Speed-based — highlight fast vs slow movement:

deltas = np.linalg.norm(np.diff(positions, axis=0), axis=1)
speed = np.concatenate([[deltas[0]], deltas])
speed_norm = (speed - speed.min()) / (speed.max() - speed.min() + 1e-8)
colors = plt.get_cmap("coolwarm")(speed_norm).astype(np.float32)

Uniform — single color for the entire trajectory:

colors = np.tile([0.2, 0.8, 0.4, 1.0], (n, 1)).astype(np.float32)

Installation

pip install trajviz
# or
uv add trajviz

Usage

import numpy as np
from trajviz import render_trajectory_video, TrajectoryRenderConfig

# Random 3D trajectory
positions = np.random.randn(120, 3).astype(np.float32)
positions = np.cumsum(positions * 0.1, axis=0)

# RGBA colors per frame (e.g., gradient from blue to red)
t = np.linspace(0, 1, 120)
colors = np.stack([t, np.zeros(120), 1 - t, np.ones(120)], axis=1).astype(np.float32)

render_trajectory_video(
    positions=positions,
    colors_rgba=colors,
    output_path="trajectory.mp4",
)

Configuration

All rendering options are controlled via TrajectoryRenderConfig:

from trajviz import TrajectoryRenderConfig, FrustumStyle

config = TrajectoryRenderConfig(
    width=1080,
    height=1080,
    elevation_deg=45.0,
    frustum_style=FrustumStyle.WEDGE,
    fps=30,
)

render_trajectory_video(positions, colors, "out.mp4", config=config)

Side-by-side comparison

Combine a recording with its trajectory overlay:

from trajviz import combine_videos_side_by_side

combine_videos_side_by_side(
    left_path="recording.mp4",
    right_path="trajectory.mp4",
    output_path="combined.mp4",
    fps=30,
)

Development

pip install -e ".[dev]"
pre-commit install

# Run tests
pytest tests/ -v

# Run linting
pre-commit run --all-files

Requirements

• Python >= 3.10
• ffmpeg on PATH
• numpy, opencv-python-headless, pydantic, jaxtyping, beartype, difflogtest