RCPCC 🚀

[ICRA 2025] Real-Time LiDAR Point Cloud Compression and Transmission for Resource-Constrained Robots 🤖

Authors: Yuhao Cao, Yu Wang, and Haoyao Chen
Affiliation: Networked Robotics and Systems Lab, HITSZ

---

Introduction 🌟

This repository provides an efficient, real-time LiDAR point cloud compression algorithm tailored for mechanically scanned LiDARs (e.g., Velodyne-64e). The algorithm leverages spatial structure of point clouds for surface fitting and employs Discrete Cosine Transform (DCT) to balance precision and compression ratio. It achieves blazing-fast compression speeds (e.g., 41.06ms for encoding and 11.35ms for decoding on an i7-124650H) and delivers high compression ratios.

Note: Due to the transmission framework’s heavy reliance on specific communication setups, we’ve open-sourced only the compression module.

For a deep dive, check out our paper:
📄 Real-Time LiDAR Point Cloud Compression and Transmission for Resource-Constrained Robots

---

How to Compile 🛠️

First, install the required libraries:

sudo apt update && sudo apt install -y libopencv-dev libfftw3-dev libzstd-dev libpcl-dev libboost-all-dev

Then, clone and build the code:

git clone https://github.com/HITSZ-NRSL/RCPCC.git
mkdir build && cd build
cmake ../src && make

---

How to Use 🎮

We’ve included a handy test program:

./build/example ./test_file/0000000000.bin 0

This command compresses a .bin point cloud file with compression level 0. Once launched, the program visualizes both the original and compressed point clouds side by side.

• Compression Levels: Choose between 0 and 5. Higher numbers = higher compression ratio (but lower precision).

---

Code Notes 📝

Compression Levels

In src/modules/serializer.cpp:

double quantization_dict[16][4] = {
    {0.25, 0.5, 0.1, 0.1}, 
    {0.25, 0.5, 0.2, 0.20},
    {0.25, 0.5, 0.4, 0.20}, 
    {0.5, 1.0, 0.1, 0.2},   
    {0.5, 1.0, 0.2, 0.2},   
    {1.0, 2.0, 0.4, 0.20}, 
};

These define the compression levels. Parameters represent:

• Pitch resolution
• Yaw resolution
• Surface fitting threshold
• DCT quantization step
  Larger values = higher compression, lower accuracy.

Configuration

In src/utils/config.h:

#define ROW_OFFSET 32.0f
#define COL_OFFSET 180.0f

#define VERTICAL_DEGREE (32.0f + 5.0f)
#define HORIZONTAL_DEGREE (180.0f + 180.0f)

Adjust these Field of View (FOV) parameters based on your LiDAR setup. The defaults work well for Velodyne-64e.

Tip: Parameters don’t need to match physical specs exactly, but improper settings may cause truncation (missing points) or redundant areas (lower compression efficiency).

---

Thanks 🙏

Explore raw data from KITTI for more experiments! The Velodyne data is plug-and-play—just specify the file path and compression level.

We’re deeply inspired by:
🌟 Real-Time Spatio-Temporal LiDAR Point Cloud Compression