Modernizing Rail Infrastructure Monitoring: A Scalable AI Pipeline for Drone-Based Photogrammetry

TrackSense is a leader in drone-based rail infrastructure monitoring, combining aerial imaging and AI analytics to detect track anomalies, monitor condition changes, and support predictive maintenance.

To sustain growth, the company needed a scalable, automated, and cost-optimized image processing pipeline that could handle large-scale datasets with minimal human intervention.

TrackSense engaged Quantum’s team to conduct an independent AI technology consulting on modernizing its photogrammetry workflow for rail-monitoring imagery.

The company sought to validate various technologies and tools while ensuring:

• Comparable accuracy and reliability in orthophoto outputs.
• Lower compute and licensing costs at scale.
• Full automation capability in AWS cloud environments.
• Enhanced cross-mission alignment for consistent longitudinal analytics.

Quantum’s consulting mission was to provide clear, quantitative evidence and an optimized architectural blueprint to guide TrackSense’s decision.

The consulting project included various tools and technologies examination, hands-on testing, cloud benchmarking, and architecture design.

Before testing, Quantum defined a detailed success criteria matrix that covered:

• Accuracy: Orthophoto alignment error thresholds and visual quality benchmarks.
• Performance: Processing time, memory, and GPU utilization metrics.
• Scalability: Ability to handle large dataset splits and parallel workloads.
• Automation readiness: Compatibility with S3 storage, batch scripts, and CI/CD triggers.
• Cost efficiency: Infrastructure usage and projected AWS cost per mission.

Each test was scored against these criteria, enabling quantitative and repeatable assessment across environments.

Quantum designed and executed a comprehensive consulting engagement and technical validation report focused on benchmarking, configuration analysis, and architectural design of a processing pipeline.

Quantum delivered a comprehensive technical evaluation report that provided TrackSense with verified performance data, cost calculations, and practical recommendations for future implementation.

The report outlined two possible development paths: proof of concept focused on fast feature extraction and anomaly detection, and a full-scale ODM pipeline implementation designed for automated, production-grade rail monitoring.

In addition to performance findings, Quantum proposed a practical workflow for data processing: generating a low-resolution orthophoto for the full dataset, processing 1 km splits at high resolution, and aligning them back to the base mosaic. The final recommendations also included transitioning from Metashape to an ODM-based processing pipeline, shifting from mission-based surveys to coverage-based data acquisition, and applying overviews to GeoTIFFs for faster visualization in QGIS.

While the final deployment remains at TrackSense’s discretion, the report serves as a validated technical foundation and actionable roadmap that the client can apply internally or extend through future implementation phases.