[CloudCorpRecords]

Closes #101

Overview

Docs-only change to src/components/detection/l_shape_fitting/l_shape_fitting_detector.py so a beginner can match each step of the code to an equation. No behavior change.

• Reference: added Zhang, Xu, Dong and Dolan, "Efficient L-Shape Fitting for Vehicle Detection Using Laser Scanners", IEEE IV 2017, at the top of the file.
• Class docstring: a 3-step overview (segment, search the direction, build the rectangle) and why a scanner sees an "L".
• _adaptive_range_segmentation: the distance-dependent threshold r_th = MIN_RNG_TH_M + RNG_TH_RATE * d.
• _merge_clusters: why overlapping neighborhoods are merged.
• _rotate_points: c1 = cos(theta) x + sin(theta) y, c2 = -sin(theta) x + cos(theta) y, and why the rectangle becomes axis-aligned in that frame.
• _calculate_variance_criterion: the distances to the nearest edge d1_i, d2_i, the split into E1/E2, gamma = -Var(E1) - Var(E2), and why the largest gamma is the best direction.
• _calculate_rectangle: the angle search and the four edge lines a*x + b*y = c, matching the a, b, c lists passed to Rectangle.
• Typo: _mearge_clusters -> _merge_clusters (and mearged_* locals). It was only used inside this file.
• One clarifying comment on min_cost_angle: the criterion is maximized, so it starts at -inf. I did not rename it, to keep the change docs-only.

Testing

• Related tests pass: pytest test/test_point_cloud_rectangle_fitting.py test/test_point_cloud_robustLS_rectangle_fitting.py test/test_rectangle.py -> 6 passed (headless, MPLBACKEND=Agg).
• To check "no behavior change" beyond the smoke tests, I ran _calculate_rectangle and _calculate_variance_criterion on a fixed synthetic L-shaped cluster (rotated 23 degrees, seeded noise) at three angle steps before and after the edit; the printed rectangle coefficients and criterion value are identical (the fit recovers 23 degrees).
• I did not run the whole suite (run_test_suites.sh), only the tests that import this module.

Notes, not changed here

• The search loop is np.arange(0, pi/2 - step, step), so the last angle tried is just under pi/2 - step, not pi/2. It is harmless for a quarter-turn-symmetric rectangle, but if you want the docstring to say "[0, 90 deg)" exactly, the end bound could be pi/2. I left the behavior as is.

AI assistance

This change was made with Claude Code (an AI coding agent), as CLAUDE.md anticipates; it wrote the docstrings and ran the checks above.