Robust Automated Inclination Assessment of Urban Electric Poles using Multi-Algorithmic Point Cloud Analysis and RANSAC-PCA Integration

Technical Summary: Robust Automated Inclination Assessment of Urban Electric Poles

Problem Statement
The structural integrity of urban utility poles, particularly prestressed concrete poles in Thailand's distribution networks, is critical for public safety and operational reliability. Traditional inspection methods rely on subjective visual evaluations and manual measurements using spirit levels, which are prone to human error, lack systematic digital records, and struggle to quantify small geometric variations (e.g., tilts under 5 degrees) with high precision. Furthermore, manual methods often fail to capture complex structural distortions, such as torsion, or to operate effectively in unstructured environments characterized by dense vegetation, overhead wiring, and ground occlusions. While recent advancements in UAVs, IoT sensors, and 2D deep learning (e.g., CNNs) have offered partial solutions, they often suffer from aerodynamic instability, hardware overhead, or a lack of depth perception and occlusion management.

Methodology
This research proposes a comprehensive, automated pipeline for assessing pole inclination using high-fidelity Terrestrial Laser Scanning (TLS) data. The study focuses on 38 low-voltage (400/230 V) prestressed concrete poles at Burapha University, utilizing a Trimble TX8 scanner to generate 3D point clouds. The methodology integrates three distinct computational approaches to process the segmented point cloud data:

1. Data Pre-processing: Raw point clouds are registered and filtered using CloudCompare to remove noise. The M-estimator Sample Consensus (MSAC) algorithm is employed to segment the pole geometry from the complex urban background and ground planes.
2. Algorithmic Approaches:
   • Approach 1 (LS & RANSAC): This method slices the pole data along the Z-axis (vertical) at 20 cm intervals. It applies Least Squares (LS) circle fitting to each cross-section to determine centroids, constructing a structural axis. Random Sample Consensus (RANSAC) is used iteratively to reject outliers and noise.
   • Approach 2 (PCA): Principal Component Analysis (PCA) is applied directly to the segmented dataset to extract the principal eigenvector, representing the pole's 3D center axis, without slicing.
   • Approach 3 (Hybrid PCA-Geometric Fitting): This hybrid strategy first uses PCA to identify the pole's initial trend line. Subsequent cross-sections are sliced perpendicular to this trend rather than the horizontal plane. These slices undergo optimal circular or rectangular fitting (using LS and RANSAC) to determine centroids, accommodating both circular and rectangular pole geometries.
3. Validation: The inclination angles derived from these algorithms are compared against ground-truth measurements obtained via a rotating dual-scale angle meter. Performance is evaluated using Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Square Error (RMSE).

Key Results
The comparative analysis reveals distinct performance characteristics for each approach depending on the structural condition of the pole:

• Upright Poles: For poles with minimal vertical deviation, the Hybrid PCA-Geometric Fitting (Approach 3) demonstrated the highest precision, achieving a minimal MAE of 0.51° and an RMSE of 0.76°. It provided the most reliable performance for near-vertical structures.
• Inclined/High-Variance Poles: For poles with significant leaning or complex environmental noise, Approach 1 (LS & RANSAC) proved superior in robustness. It achieved the lowest MAE (0.90°) and RMSE (1.08°) for inclined poles, whereas the Hybrid approach showed increased error (RMSE 1.21°) in these scenarios.
• PCA Limitations: The pure PCA approach (Approach 2) was computationally efficient but showed susceptibility to localized data clustering and bias toward high-density areas, leading to systematic inaccuracies in noisy or incomplete datasets.
• Structural Insights: The 3D point cloud methods successfully identified intricate geometric distortions and structural torsion that traditional manual measurements often overlooked, particularly in cases where the maximum lean occurred at unreachable edges.

Significance and Claims
The paper claims to establish a foundational framework for shifting utility pole inspection from subjective, manual assessments to quantitative, automated 3D analysis. The primary significance lies in the development of a multi-algorithmic pipeline that addresses the specific limitations of point cloud data in congested urban settings.

The authors assert that:

1. Robustness through Integration: No single algorithm is universally optimal; rather, a dynamic selection based on the pole's initial state (upright vs. inclined) or environmental context is necessary for the highest analytical reliability.
2. Digital Transformation: The proposed system creates a scalable digital framework for proactive maintenance, enabling the creation of longitudinal data records essential for predicting deterioration trends and structural fragility.
3. Future Adaptability: The study lays the groundwork for integrating lower-cost sensors (such as smartphone-based LiDAR) and automated feature classification into future asset management strategies, potentially enhancing the cost-viability of extensive infrastructure inspections.

The research concludes that while the Hybrid approach excels in precision for standard conditions, the LS & RANSAC methodology offers the necessary stability for managing high-variance datasets, collectively providing a more accurate and systematic tool for structural health monitoring than conventional techniques.