libcbdetCpp - Enhanced Chessboard Detection Library

Overview

Enhanced version of libcbdetCpp with improved chessboard detection algorithms, featuring robust corner detection, grid building, and missing corner completion.

Recent Updates (Latest)

Major Improvements

1. 极简1-D DBSCAN聚类算法

   • 替换复杂的DBSCAN实现为高效的1-D版本
   • 自动间距估计和参数优化
   • 支持行列标签的精确分配

2. 单应性矩阵 + 局部搜索预测缺失角点

   • 使用cv::findHomography建立(r,c)→(x,y)映射
   • localCornerSearch基于OpenCV棋盘格角点检测候选点
   • 完全移除像素模板匹配，提高鲁棒性
   • 支持三种点类型：0=预测-低置信, 1=原测得, 2=补全

3. 改进的角点补全系统

   • predictMissingCorners: 基于单应性矩阵的预测
   • completeMissingCorners: 简化的补全流程
   • 智能权重处理，对补全点使用较低权重

4. 增强的异常值剔除

   • rejectOutliers2: 支持点类型权重
   • 自适应k_sigma搜索
   • 局部一致性检查

Key Features

• Robust Corner Detection: Enhanced libcbdetect integration
• Dynamic Parameter Adjustment: Automatic parameter optimization
• Grid Quality Validation: Fill rate and spacing analysis
• Bundle Adjustment Support: Optional Ceres Solver integration
• Comprehensive Visualization: Grid drawing and result analysis

Build Instructions

mkdir build && cd build
cmake ..
make

Usage

./bin/example <image_path>

Dependencies

• OpenCV 4.x
• Optional: Ceres Solver (for bundle adjustment)

Test Results

• Color.bmp: 81→119角点，填充率63.6%
• IR.bmp: 95→112角点，填充率59.1%

Technical Details

DBSCAN Clustering

void dbscan1D(const std::vector<float>& proj, float eps, int min_samples, std::vector<int>& labels)

• 极简1-D实现，基于排序的区间扩张
• 自动标签归一化
• 参数：eps = 0.3 * median_spacing, min_samples = 2

Corner Completion

void predictMissingCorners(const cv::Mat& gray, ...)
void completeMissingCorners(const cv::Mat& gray, ...)

• 单应性矩阵预测
• 局部搜索精化
• 智能权重处理

Outlier Rejection

void rejectOutliers2(..., const std::vector<char>& point_types, float k_sigma = 3.0f)

• 支持点类型权重
• 自适应阈值搜索
• 局部一致性验证

License

Original libcbdetect license applies.