PyData Yerevan 2022

Using Few Shot Object Detection for Utility Pole detection from Google Street View images.
08-13, 13:45–14:25 (Asia/Yerevan), 213W PAB

Traditional methods of detecting and mapping utility poles are manual, time-consuming and costly processes. Current solutions focus on detection of T-shaped (cross-arm shaped poles) and the lack of labeled data makes it difficult to generalize the process of other types of poles. This work aims to use Few Shot Object Detection techniques to overcome the unavailability of the data and to create a general pole detection model with few labeled images.


Utility poles play an essential role for wiring neighborhoods with electricity and telecommunication supply. The absence of aerial poles creates additional work for installing the wires underground, which is a much more time-consuming and costly process. Identification of neighborhoods with available utility poles helps the fiber, wireless and mobile broadband market providers to make decisions on market expansion strategies. Traditional methods of detecting and mapping utility poles are manual, time-consuming and costly because of the demand for visual interpretation with quality data sources or intense field inspection. Existing works rely on the detection of T-shaped (cross-arm shaped) utility poles, however other types of poles (straight poles, light poles) are not being identified. Around 50% of poles around US neighborhoods are straight poles. Additionally, there is a lack of quality-labeled datasets of any types of poles.

This works aims to observe different directions of generalizability of high-quality T pole detection models to other types of poles, specifically, we handled that with different approaches including Few Shot Object Detection.


Prior Knowledge Expected

Previous knowledge expected

Mark Hamazaspyan is a Machine Learning specialist with 3+ years of experience in computer vision, 4+ years of experience in traditional data science, and 2+ experience in ML engineering and MlOps. He has completed various projects, including computer vision, time series, and event data. Mark Hamazaspyan has a teaching experience in universities and private organizations.

Education:
2021-present - Applied Statistics and Data Science, Yerevan State University.
2017-2021 - BA in Business: Economics, American University of Armenia