CV Detectors

GoodFeaturesToTrack (Shi-Tomasi)

goodFeaturesToTrack finds corner points and high-contrast features. In tracking pipelines those points are followed across frames (e.g. Lucas–Kanade optical flow) and rendered as motion trajectories — between machine vision and glitch / data aesthetics.

• maxCorners — upper limit on returned points
• qualityLevel — relative corner quality threshold
• minDistance — minimum spacing between points

HoughLinesP

cv::HoughLinesP is the Probabilistic Hough Transform: it returns finite line segments (x1,y1,x2,y2) instead of infinite lines. Useful for lane-like structure, document edges, and reducing a scene to its linear skeleton for generative drawing.

• threshold — votes needed to accept a line
• minLineLength — discard short segments
• maxLineGap — merge nearby collinear fragments

HoughCircles

cv::HoughCircles detects circles via the gradient-based Hough Circle Transform. It is fast when edges are clean, but sensitive to parameter tuning — round objects in footage become centers and radii for geometry, instancing, or interactive control.

• minDist — spacing between circle centers
• param1 / param2 — Canny high threshold and accumulator threshold
• minRadius / maxRadius — search range

Scene structures

Built on detected lines: Manhattan (vertical/horizontal) filtering, vanishing-point estimates, grid hints, and quadrilateral regions — a lightweight way to label architectural cues in the frame for study or overlay aesthetics.