/jetson-stereo-depth

Primary LanguagePythonMIT LicenseMIT

Jetson Stereo Depth

Build your own HW-accelerated stereo camera from scratch with Nvidia Jetson. demo gif

This repository contains:

  1. Stereo Pair intrinsic calibration

image

2. Lens distortion correction

image

3. VPI HW-accelerated `remap()`

image

  1. Depth calculation demo gif

  2. 3D Printable Hardware IMG_4761

  • HW-accelerated depth pipeline (capture->rectify->SGBM->depth calc.) in python
  • Reference disparity pipeline with OpenCV CPU implementation
  • Calibration notebooks to determine a) intrinsic parameters and lens distortion coeficients of cameras b) rectification map between two sensors
  • Deployment code in C++
  • Assembly instructions for the hardware

Getting started

  1. Install dependencies
sudo apt install libnvvpi1 python3-vpi1 vpi1-dev

  1. Calibrate cameras (see notebooks in calib/ folder).

  2. Try python implementation

cd depth_pipeline_python
python3 depth_vpi.py
  1. Try C++ implementation
cd depth_pipeline_cpp
mkdir build && cd build
cmake .. && make -j$(nproc)
./depth_pipeline

Pipeline

depth_pipeline

Hardware

3D printable holder for 2 × IMX219 camera module is available here. Baseline distance of the pair is 80mm.