Probabilistic pipe material classification via LiDAR-IMU data fusion

Reliable identification of pipe material in underground sewer networks is essential for targeted maintenance and asset management. This paper presents a probabilistic framework for classifying clay, concrete, and plastic pipes by fusing LiDAR intensity with IMU-based joint detection. LiDAR intensity is normalised for range and scan angle effects using a shared bivariate polynomial model fitted across all materials, producing residuals that are sensitive to material type and whose distributions are captured by kernel density estimation (KDE). Independently, the Inertial Measurement Unit (IMU) detects mechanical perturbations at pipe joints. The gaps between detected joints are compared with the expected segment length for each material. Long gaps can be interpreted as missed-joint cases when they are close to integer multiples of the expected segment length, while unreliable gaps are down-weighted by an outlier component. Both modalities share a unified Bayesian framework with uniform priors, and are combined in the log domain using mean log-likelihoods with an adjustable modality weight. Experiments use six controlled runs across three single-material pipe networks with distinct segment lengths. In this setting, intensity alone achieves posterior confidence of approximately 49-60%, while distance alone reaches 69-86.5%. Fusion increases posterior confidence relative to either single modality in all cases, assigning 93.5%, 93.5%, and 86.2% posterior probability to the true concrete, clay, and plastic classes, respectively.