Stereo Capture Tool

DepthAI capture script for stereo (and optionally RGB/depth) streams from OAK devices. Saves frames as .npy files with timestamps and optional calibration/metadata.

Requirements

• DepthAI-compatible device (OAK camera) connected via USB or network
• Python 3.9+

Setup

Create Virtual Environment

Option 1: Using Conda

conda create -n capture python=3.11
conda activate capture

Option 2: Using Python venv

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

Install Requirements

pip install -r requirements.txt

Usage

python capture_data_stereo.py [OPTIONS]

Options

Option	Description
--settings	Path to settings JSON (default: capture_settings.json)
--output	Custom output root folder (default: output next to the script)
--capture-name	Name for the capture (included in folder name and metadata)
--ip	Device IP for network connection (omit for USB)
--autostart	Start capturing after N seconds (-1 = disabled, default)
--autostart_time	Start at a fixed datetime (e.g. from cron)
--autostart_end	Stop at a fixed datetime
--no-streams	Do not show stream windows; use control window for S/Q (faster capture)
--png	Save left, right, rgb as PNG (disables npy unless --npy is also set)
--npy	Save frames as numpy (default when no format option is set). Use with --png to save both.

Captures are saved under the output folder in subfolders named by device, optional capture name, and timestamp (e.g. output/OAK-D_abc123_myrun_20250211120000/). By default each stream is saved as {stream}_{timestamp_ms}.npy. With --png, left/right/rgb are saved as .png; with both --png and --npy, those streams are saved in both formats. Calibration and metadata are written in the same folder.

Settings

Edit capture_settings.json to configure the pipeline and capture.

• num_captures: Max frames per stream (20, or "inf" for unlimited).
• output_settings: Enable/disable streams: left, right, left_raw, right_raw, rgb, depth, disparity; select sync for synchronized capture.
• stereoResolution / rgbResolution: {"x": width, "y": height}.
• ir / ir_value: IR laser dot projector (0–1).
• flood_light / flood_light_intensity: IR flood light.
• FPS: Target FPS.

Higher FPS (disable stream display)

To reduce overhead and achieve higher capture FPS, disable the live stream windows and use only the small control window:

python capture_data_stereo.py --no-streams

Controls

• s: Start or stop capture
• q: Quit