YOLO Application on Portuguese Highways for Photovoltaic Energy Potential Evaluation

The global push for renewable energy has accelerated the deployment of photovoltaic (PV) systems, raising concerns about land availability and competition with other land uses. A promising solution lies in repurposing underutilised highway infrastructure, such as sound barriers and excavation slopes, for PV installations. This thesis investigates the application of computer vision, specifically the YOLOv10 object detection algorithm, to automatically identify these structures along highways using video and image data.

The methodology proposed in this work, which combines object detection with geolocation, enables the estimation of electric energy generation and supports the technical-economic assessment of the viability of these infrastructures for renewable energy production.

For sound barriers, several YOLOv10 models were trained on progressively larger datasets, improving mAP@0.5 from 0.57 to 0.90. The geolocation method developed, based on the synchronisation of GoPro videos with GPS data, makes it possible to map detected barriers and estimate their solar potential. In a case study involving 64 sound barriers, two PV installation scenarios were evaluated, achieving energy yields of 17.8 kWh/kW and 27.7 kWh/kW per barrier. These results compare favourably with similar projects in Europe, confirming the strong solar potential in Portugal.

Excavation slopes are visually more complex structures and harder to identify. Despite using a modest dataset, the model was still able to successfully detect these types of infrastructure, achieving a mAP@0.5 of 0.54. The case study showed that these slopes could generate between 88 and 170 kWh/year per m2, depending on their orientation.