PY_SPRT_RANSAC:R-RANSAC with Sequential Probability Ratio Test

This project implements several variants of the SPRT (Sequential Probability Ratio Test) RANSAC algorithm for robust plane fitting in 3D point clouds. The algorithms are optimized for speed(using numba) and accuracy, and support both standard and normal-assisted sampling strategies.

Based on Matas and Chum's 2008 paper link and fiskrt's Matlab implementation link. (But have some differents and new features)

1.Features:

four variants of the SPRT-RANSAC algorithm:

• SPRT-RANSAC : Standard SPRT-RANSAC for robust plane fitting
• SPRT-RANSAC-with-Normal: Normal-assisted SPRT-RANSAC for scenarios with known point normals
• Fast SPRT-RANSAC&SPRT-RANSAC-with-Normal: Fast approximate versions of both SPRT-RANSAC and SPRT-RANSAC-with-Normal, but accuracy may be lower.

Speed up by Numba.

2.Main Functions

Funtion: SPRT_RANSAC_RAW

Fits a plane to 3D points using standard SPRT-RANSAC with three-point random sampling.

Parameters

• points: (N, 3) array of 3D points
• threshold: Distance threshold for inliers
• eta0, epsilon, delta: SPRT parameters
• max_iterations, m, tm, ms: Algorithm parameters

Returns

• best_model: Plane coefficients [A, B, C, D] (Ax + By + Cz + D = 0)
• best_support: Number of inliers
• mean_distance_error: Mean absolute distance to the plane

SPRT_RANSAC_RAW_NOR

SPRT-RANSAC using point normals for model estimation (no random sampling).

Parameters

• points: (N, 3) array of 3D points
• point_normals: (N, 3) array of normals
• eta0, epsilon, delta: SPRT parameters
• max_iterations, m, tm, ms: Algorithm parameters

Returns

• best_model: Plane coefficients [A, B, C, D] (Ax + By + Cz + D = 0)
• best_support: Number of inliers
• mean_distance_error: Mean absolute distance to the plane

SPRT_RANSAC_FAST

A faster, approximate SPRT_RANSAC_RAW version. Parameters/Returns: Same as SPRT_RANSAC_RAW

SPRT_RANSAC_FAST_NOR

A faster, approximate SPRT_RANSAC_RAW_NOR version.

Parameters/Returns: Same as SPRT_RANSAC_RAW_NOR

Dependencies

numpy numba

Example Usage

import numpy as np
from RRANSAC import SPRT_RANSAC_RAW, SPRT_RANSAC_RAW_NOR, SPRT_RANSAC_FAST, SPRT_RANSAC_FAST_NOR

points = np.random.rand(1000, 3).astype(np.float32)
normals = np.random.rand(1000, 3).astype(np.float32)

model, support, mean_err = SPRT_RANSAC_RAW(points)
print("Best plane:", model)
print("Inliers:", support)
print("Mean error:", mean_err)

model_nor, support_nor, mean_err_nor = SPRT_RANSAC_RAW_NOR(points, normals)
print("Best plane with normals:", model_nor)
print("Inliers with normals:", support_nor)
print("Mean error with normals:", mean_err_nor)

model_fast, support_fast, mean_err_fast = SPRT_RANSAC_FAST(points)
print("Best plane (fast):", model_fast)
print("Inliers (fast):", support_fast)
print("Mean error (fast):", mean_err_fast)

model_fast_nor, support_fast_nor, mean_err_fast_nor = SPRT_RANSAC_FAST_NOR(points, normals)
print("Best plane with normals (fast):", model_fast_nor)
print("Inliers with normals (fast):", support_fast_nor)
print("Mean error with normals (fast):", mean_err_fast_nor)