Pothole Detection and Material Estimation: Driving Efficiency in Road Infrastructure Management
Summary
Road maintenance agencies face a significant challenge in the timely and cost-effective repair of potholes. Traditional methods are slow, labor-intensive, and often lack the precision needed for accurate resource allocation. Our solution leverages a novel, AI-driven framework to automate the identification of road defects and accurately estimate the required repair materials. By combining YOLOv8 segmentation, depth estimation, and a geo-tagging system, we have developed an end-to-end application that provides real-time, actionable insights for proactive and efficient road repair. The implementation has led to a quantifiable reduction in operational costs and a significant improvement in maintenance workflow.
The Challenge
Aging infrastructure and increasing traffic volumes have made pothole management a critical concern for municipal and state transportation departments. Key issues include:
- Inefficient Detection: Manual inspection is time-consuming, subjective, and exposes workers to road hazards.
- Inaccurate Material Estimation: Repairs often use a “best-guess” approach, leading to material waste or insufficient supplies, which increases costs and delays project completion.
- Reactive Maintenance: Potholes are typically repaired only after they’ve been reported, by which time they have likely worsened and become more hazardous.
- Lack of Data-Driven Planning: Without accurate data on the size, depth, and location of potholes, it’s impossible to create optimized repair schedules or allocate budgets effectively.
Our client needed a solution that could transform their reactive, manual process into a proactive, data-informed strategy. They required a system that could not only detect potholes but also provide the precise measurements needed to streamline their repair operations.
Our Solution
We developed a comprehensive software application that uses computer vision and deep learning to address the client’s core challenges. The solution, built on a Streamlit framework for a user-friendly interface, integrates three core technologies:
- High-Accuracy Pothole Segmentation
We trained and deployed a YOLOv8-seg model (best.pt) on a custom dataset of road images. The model’s segmentation capabilities allow it to precisely outline the boundaries of each pothole, distinguishing it from the surrounding road surface with high accuracy. The model achieved excellent performance, with a precision of 95.2% and an Intersection over Union (IoU) of 88.5%, significantly outperforming traditional object detection models. - Depth Estimation for Volumetric Analysis
To move beyond simple 2D area measurements, we integrated a pre-trained depth estimation model (Intel/dpt-large or Intel/dpt-hybrid-midas). This model analyzes the image to create a depth map, where darker pixels represent deeper points. By combining the segmentation mask with the depth map, our system can calculate the median depth and, most importantly, the volume of each pothole in cubic meters. This is a critical factor for accurate material estimation. - Material and Cost Calculation
Using the calculated volume, the system applies a set of predefined material constants to estimate the exact resources required. This includes:- Compaction Factor: A multiplier (DEFAULT_COMPACTION_FACTOR = 1.25) is applied to account for material settling.
- Asphalt Density: The system uses a standard asphalt density (DEFAULT_ASPHALT_DENSITY_KG_M3 = 2300.0) to convert volume into mass (kg).
- Resource Quantification: The mass is then used to determine the number of asphalt bags and the total tonnage required.
- Cost Projection: A cost-per-bag metric (DEFAULT_COST_PER_BAG = 300.0) provides an immediate, on-site cost estimate for each repair.
- Automated Geo-Tagging and Data Reporting
The application automatically extracts GPS coordinates from image EXIF data or uses IP-based fallback locations, linking each pothole detection to a precise geographical point. All data—including area, depth, volume, mass, cost, and location—is compiled into a downloadable CSV report, enabling data-driven decision-making and project planning.
Results and Impact
The implementation of this technology has yielded powerful, tangible results:
- Enhanced Efficiency & Cost Reduction
By shifting from manual, reactive processes to an automated, proactive system, the client achieved a 25% reduction in labor hours for road inspections. The precise material estimation, validated through several pilot projects, led to a 15% decrease in material waste, resulting in significant cost savings. - Proactive & Optimized Planning
The detailed, location-specific reports enabled the client to move from fixing one-off reports to creating optimized repair routes. Teams can now be dispatched with the exact amount of material needed, eliminating return trips and minimizing downtime. - Improved Safety
Automating the detection process keeps maintenance staff off the road and out of harm’s way, reducing the risk of accidents.
The technology demonstrated superior performance compared to an alternative model that only provided bounding box detection. Our segmentation model provided a more accurate representation of the pothole’s irregular shape, which, when combined with volumetric calculations, yielded a much more reliable material estimate. This comprehensive approach is a true paradigm shift in road maintenance.
Comparison: Our Approach vs. Bounding Box-Only Detection
Feature | Our Segmentation + Depth Approach | Bounding Box-Only Detection |
Pothole Shape | Precisely outlines irregular shapes with a pixel-level mask. | Approximates the shape with a rectangular box, including surrounding road. |
Material Estimation | Highly Accurate. Calculates volume based on segmented area and estimated depth. | Inaccurate. Only provides a rough area estimate (L × W) without depth, leading to waste. |
Data Quality | Rich, detailed data including area, median depth, and volume. | Limited to basic metrics like bounding box coordinates and confidence scores. |
Operational Impact | Enables proactive, data-driven resource allocation and route planning. | Provides a simple count of potholes, which is less useful for planning. |
Cost Savings | Direct cost savings from reduced material waste and improved labor efficiency. | Limited to no direct cost savings. |
