top of page
Yazarın fotoğrafıBusra Demir

Satellite Image Vehicle Data Labeling with Hamad Bin Khalifa University (HBKU) - Qatar Computing Research Institute (QCRI) & Co-one

Overview


Satellite Image Vehicle Data Labeling

Co-one partnered with Qatar Computing Research Institute (QCRI), a part of Hamad Bin Khalifa University (HBKU), to work on an innovative project in space technologies, focusing on urban surveillance using satellite imagery. The project involved automated data annotation for satellite images, specifically targeting vehicles such as cars, trucks, and buses. Co-one provided advanced oriented bounding box annotation, enhancing the precision and usability of vehicle detection in satellite data.


The Problem


Labeling objects in satellite images presents unique challenges due to the distance from which the images are captured, making it difficult to identify and annotate vehicles accurately. The orientation and scale of vehicles (cars, trucks, buses) in such images required rotation of the bounding boxes to match the objects correctly. This project demanded high accuracy to ensure the data generated would be reliable for urban surveillance applications.


The Solution


As Co-one, we leveraged our automated data annotation platform to handle the complexity of labeling vehicles in satellite images. Using advanced oriented bounding boxes, we rotated the boxes to accurately fit each vehicle—whether it was a car, truck, or bus—despite the challenges posed by the distance and angle of the images. Our team meticulously processed 4,300 satellite images, resulting in more than 100,000 precise labels.


Rotated Bounding Box Labeling
Rotated Bounding Box Labeling
Unrotated Bounding Box Labeling
Unrotated Bounding Box Labeling

As Co-one, we rapidly completed a process that would have taken months in just days—achieving results 10 times faster than if done in-house. Our Solution Architects identified and flagged edge cases within the dataset, ensuring that no critical scenarios were overlooked. Additionally, we helped prevent bias by thoroughly analyzing the dataset and reporting missing labels, enabling a highly accurate dataset.


By applying this technique, we ensured that every label was correctly aligned, contributing to a high accuracy rate for this challenging dataset.


The Result


Through collaboration with HBKU - QCRI, we successfully annotated vehicles across thousands of satellite images, supporting the accuracy of urban surveillance data for space technologies. The advanced oriented bounding box technique ensured that even difficult-to-label objects, such as vehicles viewed from high altitudes, were accurately annotated, providing valuable data for further AI-based analysis and surveillance.


Thanks to our rapid and precise data annotation approach, we significantly reduced project completion time while enhancing data quality.


Customer Review


High-quality data is a key driver of success in today’s tech industry, and it was exactly what we needed for our vehicle detection project using satellite images. Partnering with Co-one on this initiative was a great decision. Their meticulous attention to detail and ability to address complex edge cases resulted in a reliable dataset with precise oriented bounding box annotations. Leveraging this dataset, we successfully trained state-of-the-art vehicle detection models, which we plan to use in applications like traffic management, urban planning, and disaster response. Co-one’s expertise and efficiency not only ensured exceptional data quality but also significantly reduced our project turnaround time, making them an invaluable partner in our research efforts.”

Dr. Ferda Ofli

bottom of page