Pothole Detection Using MATLAB Computer Vision and Deep Learning

A student project that finds potholes in top-down road images, measures them and turns them into a repair plan. Four detectors are compared on 400 held-out images with 513 potholes: a classical image-processing chain (object F1 0.487), HOG + LBP + SVM (0.191), a U-Net trained from scratch (0.929) and an LR-ASPP MobileNetV3 trained by transfer learning (0.975). Each pothole is measured (area, size, estimated depth) and graded with the ASTM D6433 severity table, and a repair estimate becomes a work order. A drive-through survey runs the same steps on roof-camera video: bird's-eye view, multi-frame tracking, chainage and GPS, 25 m segment ratings. A role-based web dashboard brings it together. Honest scope: MATLAB is NOT installed on the build PC - the MATLAB scripts were written but never run, and every number comes from the Python implementation; all images are synthetic (procedural rendering, no real photographs, no public dataset), so the survey's 34/34 result is an upper bound.

Category: Civil Engineering / Transportation and Highway Engineering / Pavement Condition Assessment / Computer Vision / Deep Learning / Road Asset Management / MATLAB (scripts) with Python

Civil Engineering Project C07 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.