PCD to Occupancy Grid Map Conversion

This project converts 3D LiDAR point clouds (.pcd) into 2D occupancy grid maps (.pgm + .yaml) compatible with the ROS navigation stack.

The pipeline performs ground removal, obstacle extraction, grid projection, and map post-processing to generate a clean 2D map suitable for localization and navigation.

---

Pipeline Overview

PCD → Ground Filtering → 2D Occupancy Grid Projection → Map Image (.pgm) → YAML Metadata → Post-Processing → Clean Navigation Map

Two ground filtering approaches are implemented:

1. RANSAC-based ground removal
2. Grid-based ground filtering

Both methods convert the filtered obstacle points into a 2D occupancy grid.

---

1. RANSAC-Based Map Generation

This approach removes the ground plane using the RANSAC plane segmentation algorithm.

Method

1. Load the .pcd file using Open3D.
2. Remove statistical outliers to reduce LiDAR noise.
3. Use RANSAC plane segmentation to detect the dominant plane (floor).
4. Remove ground points.
5. Project remaining obstacle points onto a 2D XY grid.
6. Convert the grid into an occupancy map.
7. Apply Gaussian blur and dilation to improve map continuity.
8. Save the final map as .pgm and .yaml.

Key Parameters Used

Parameter	Value	Purpose
nb_neighbors	20	Used in statistical outlier removal
std_ratio	2.0	Removes noisy points
distance_threshold	0.02	RANSAC plane fitting tolerance
ransac_n	3	Number of points sampled per iteration
num_iterations	1000	Improves plane detection robustness
resolution	0.05 m	Grid cell resolution
GaussianBlur kernel	5×5	Smooths sparse obstacles
dilation kernel	3×3	Expands obstacle regions

Why These Parameters Improve the Map

• Statistical outlier removal reduces random LiDAR noise.
• RANSAC segmentation removes the floor plane effectively.
• Gaussian blur fills small gaps in obstacle clusters.
• Dilation strengthens obstacle boundaries for navigation safety.

---

2. Grid-Based Ground Filtering

Instead of fitting a global plane like RANSAC, this approach estimates ground height locally for each grid cell.

This method works better when the ground surface is uneven or sloped.

Method

1. Load the .pcd file.
2. Apply statistical outlier removal.
3. Divide the XY space into grid cells.
4. Store the minimum Z value per grid cell as the ground estimate.
5. Classify points as obstacles if they lie above the ground threshold.
6. Convert obstacle points into a 2D occupancy grid.
7. Apply smoothing and dilation.
8. Save the resulting map.

Key Parameters Used

Parameter	Value	Purpose
ground_threshold	0.15 m	Minimum height above ground to classify obstacle
max_height	5.0 m	Upper limit for obstacle height
resolution	0.05 m	Grid map resolution
GaussianBlur kernel	3×3	Smooth obstacle clusters
dilation kernel	5×5	Strengthens obstacle representation

Why These Parameters Improve the Map

• Local ground estimation handles uneven floors better than RANSAC.
• Ground threshold filtering removes floor reflections.
• Height filtering avoids ceiling points or tall noise.
• Larger dilation kernel improves wall connectivity in the final map.

---

3. Map Post-Processing

After generating the occupancy grid, small noisy clusters may still exist.

Post-processing performs:

1. Binary obstacle mask generation
2. Connected component analysis
3. Removal of small obstacle clusters
4. Reconstruction of a clean occupancy grid

This produces a smoother map suitable for robot navigation.

---

Map Output Format

The final map is saved in the standard ROS map format:

PGM Image

Represents occupancy information:

Value	Meaning
0	Occupied
254	Free
205	Unknown

YAML Metadata

Example:

image: map.pgm
resolution: 0.05
origin: [x, y, 0.0]
negate: 0
occupied_thresh: 0.65
free_thresh: 0.196

This YAML file is used by the ROS map_server node.

---

Results

For evaluation:

• Screenshots of generated PGM maps were taken directly.

• Each method was tested with tuned parameters.

• Maps were compared based on:

  • obstacle continuity
  • noise reduction
  • wall clarity
  • navigable space quality

The tuned parameters improved map quality by:

• reducing ground artifacts
• eliminating small noise clusters
• improving obstacle boundary definition

(Insert PGM screenshots here)

---

How to Run the Code

Requirements

ROS Noetic Python3 Open3D NumPy OpenCV

Install dependencies:

pip install open3d opencv-python numpy

---

Running the RANSAC Map Generator

rosrun <package_name> ransac_map_generator.py <input_pcd_file>

Example

rosrun pcd_to_map ransac_map_generator.py map_radius_filter.pcd

Input

map_radius_filter.pcd

3D LiDAR point cloud file.

Output

map_name.pgm
map_name.yaml

These files represent the generated occupancy map.

Optionally the map can be launched using ROS:

map_server map_name.yaml

---

Running the Grid-Based Map Generator

rosrun <package_name> grid_map_generator.py

Input

pcd_files/map_radius_filter.pcd

Output

• Occupancy Grid published to /map
• Saved .pgm map
• Saved .yaml metadata

The map can be visualized in RViz.

---

ROS Topic Output

Published Topic:

/map

Message Type:

nav_msgs/OccupancyGrid

Frame:

map

---

Summary

This project provides two different approaches for generating 2D occupancy maps from 3D LiDAR data:

Method	Strength
RANSAC	Best for flat environments
Grid-Based	Handles uneven ground surfaces

Both methods generate ROS-compatible navigation maps and allow parameter tuning to improve map quality.

---

PCD to PNG Map Conversion Results

This table compares the output maps generated from the same input PCD using two different approaches:

• GRID Mapping Algorithm
• RANSAC Algorithm

Each result is shown after post-processing (MPP) for both algorithms.

Input PCD	GRID + MPP	PTM + MPP
cat_adc.pcd
cat_adc_t2.pcd
dematic.pcd
dematic2.pcd
dematic3.pcd
h2l2.pcd
h2l2_combine_new.pcd
h2l2_downloaded.pcd
map_filter.pcd
map_filter_ptm.pcd
map_radius_filter.pcd