Drone-based solar inspection

Summary

Computer vision for panel inspection and reporting.

Inspection needed to identify panels in drone video, choose useful viewing angles and link suspected faults to locations.

A segmentation and tracking pipeline using thermal/RGB video, with TensorRT optimisation and a web service for inspection reports and map exports.

Tech Challenge

• Real-time binary segmentation of solar panels from the video.
• Choosing the "moment" of the panel classification at the right angle.
• Mapping the corresponding panel to its position on the map.
• Creation of interactive report with different view parameters.
• Creation of corresponding .kmz file with coordinates mapping.
• Creation of energy and money loss analytics based on classification results.

Solution

• Segmentation and tracking of panels based on TRGB video data.
• Optimization of processing with TensorRT.
• Using U-net for binary segmentation.

Outcome

A service for recurring inspections and uploaded-video analysis, producing mapped findings and estimated loss reports.

Users can analyse uploaded drone video and receive mapped findings, KMZ exports and estimates of energy and financial losses.