| Limitation | Conventional Method | 360° LIDAR Approach | |------------|--------------------|---------------------| | | Limited to line‑of‑sight, often requires multiple passes | Whole‑scene capture in a single scan | | Subjectivity | Dependent on inspector experience | Objective geometric measurements | | Speed | Hours to days for large structures | Minutes per scan (≤ 5 min) | | Environmental constraints | Poor lighting, weather sensitivity | Independent of illumination; works in low‑light | | Data richness | 2‑D images only | 3‑D geometry + intensity + RGB (when fused) |
360° LIDAR‑Based Crack Detection and Characterisation for Infrastructure Inspection lidar360 crack
LIDAR, 360° scanning, crack detection, point‑cloud processing, infrastructure inspection, deep learning, non‑destructive evaluation. 1. Introduction Crack detection is a cornerstone of structural health monitoring (SHM) for civil engineering assets. Early identification enables timely maintenance, extending service life and preventing catastrophic failures. Traditional inspection techniques—handheld visual surveys, hammer probing, or 2‑D photogrammetry—suffer from several drawbacks: | Limitation | Conventional Method | 360° LIDAR
A. Smith¹, B. Lee², C. Martínez³, D. Khan⁴ Lee², C
A. Smith (asmith@univx.edu) Abstract Crack formation is a primary indicator of structural deterioration in concrete, asphalt, and rock surfaces. Conventional visual inspection is labor‑intensive, subjective, and limited to line‑of‑sight. This paper presents a fully automated 360° LIDAR‑based crack detection (LIDAR‑360‑Crack) pipeline that exploits high‑resolution terrestrial laser scanning (TLS) to acquire dense point clouds of entire structural façades, bridges, tunnels and pavements in a single sweep. By integrating multi‑scale geometric descriptors, intensity‑based filtering, and a lightweight deep‑learning classifier, the system extracts crack geometries, quantifies their width, depth and orientation, and generates GIS‑compatible vector maps. Extensive field trials on three bridge decks, two highway sections and a historic stone wall demonstrate detection accuracies of 94.2 % (precision) / 91.8 % (recall) , with mean absolute width error < 0.4 mm. The proposed framework reduces on‑site inspection time by 70 % relative to manual methods and offers a reproducible dataset for long‑term structural health monitoring.