| Abstract |
Road maintenance is essential for ensuring transportation safety, efficiency and sustainability.
Traditional road inspection methods relied heavily on human labour. Recently, intelligent
approaches that utilize remote sensing technology, such as LiDAR and cameras have gained
momentum, significantly reducing human effort. However, the collected data is often manually
processed, which can be time consuming and inefficient. Automating road maintenance by
leveraging advanced artificial intelligence (AI) and deep learning techniques offers a transformative
solution to this challenge. Nonetheless, deep learning models for road object detection require large
annotated datasets for training, which are labour-intensive and time consuming to create. They also
tend to have limited generalization abilities for unseen objects. To overcome this issue, we propose
a novel approach that uses meta learning algorithms and data augmentation to enhance the
performance of automated road asset detection from spherical images. Meta learning enables
models to generalize effectively from limited labelled data, reducing the dependency on large
datasets. Our approach leverages the advanced object detection capabilities of deep learning
models, while eliminating the effort required to prepare large training datasets. This allows the
model to be easily re-trained for different scenarios. Notably, our method has been proven to work
effectively on data from various cities, demonstrating its robustness across diverse road
environments. This presentation provides insights into the details of using AI technology to improve
the efficiency of processing spherical imagery for road maintenance applications and opens up
opportunities for the large-scale deployment within cities.
|
