SODA: Site Object Detection dAtaset for Deep Learning in Construction (2202.09554v1)

Abstract: Computer vision-based deep learning object detection algorithms have been developed sufficiently powerful to support the ability to recognize various objects. Although there are currently general datasets for object detection, there is still a lack of large-scale, open-source dataset for the construction industry, which limits the developments of object detection algorithms as they tend to be data-hungry. Therefore, this paper develops a new large-scale image dataset specifically collected and annotated for the construction site, called Site Object Detection dAtaset (SODA), which contains 15 kinds of object classes categorized by workers, materials, machines, and layout. Firstly, more than 20,000 images were collected from multiple construction sites in different site conditions, weather conditions, and construction phases, which covered different angles and perspectives. After careful screening and processing, 19,846 images including 286,201 objects were then obtained and annotated with labels in accordance with predefined categories. Statistical analysis shows that the developed dataset is advantageous in terms of diversity and volume. Further evaluation with two widely-adopted object detection algorithms based on deep learning (YOLO v3/ YOLO v4) also illustrates the feasibility of the dataset for typical construction scenarios, achieving a maximum mAP of 81.47%. In this manner, this research contributes a large-scale image dataset for the development of deep learning-based object detection methods in the construction industry and sets up a performance benchmark for further evaluation of corresponding algorithms in this area.

• The paper presents SODA, a large-scale, annotated dataset designed to boost deep learning object detection within construction environments.
• It details meticulous image collection and annotation processes across 15 object classes, ensuring high-quality, precise bounding boxes.
• Benchmarking with YOLO v3 and YOLO v4 shows impressive performance, setting a new standard for automated construction site analysis.

The paper "SODA: Site Object Detection dAtaset for Deep Learning in Construction" by Rui Duan et al. introduces Site Object Detection dAtaset (SODA), an expansive dataset specifically designed for advancing deep learning object detection within the construction industry. The SODA dataset comprises more than 19,000 images, featuring 286,201 objects across 15 distinct categories, which are classified into four major groups: workers, materials, machines, and layout elements.

Motivation and Need

The motivation for constructing the SODA dataset arises from the lack of large-scale, open-source datasets tailored specifically to the construction industry. While computer vision-based deep learning algorithms have seen substantial development, leveraging these techniques in construction is hindered by the absence of domain-specific annotated image datasets. This deficiency limits the growth and application of deep learning models in construction site management tasks such as safety monitoring and productivity analysis.

Dataset Composition and Collection

• Image Collection: The dataset was compiled from 20,000+ images obtained from various construction sites in Guangzhou, China. The data collection involved multiple equipment types, including monocular cameras, UAVs, and construction site monitoring systems, capturing images from different angles, lighting conditions, and construction phases.
• Categories: The dataset's 15 object classes comprise categories like workers (person, helmet, vest), materials (board, wood, rebar, brick, scaffold), machines (handcart, cutter, electric box, hopper, hook), and layout (fence, slogan). This categorization reflects the critical components encountered on construction sites.

Data Processing and Annotation

The dataset underwent rigorous data cleaning and annotation processes:

• Data Cleaning: Involved eliminating duplicate, ambiguous, and irrelevant images, alongside privacy measures such as anonymizing identifiable company logos and human features.
• Annotation: Implemented following stringent guidelines to ensure precise bounding boxes around objects, conducted using VOC format. Thirty-five civil engineering students contributed significantly to the annotation task under expert supervision, ensuring the dataset's reliability and accuracy.

Statistical Analysis and Benchmarking

The dataset demonstrated a robust diversity and comprehensiveness in its category coverage:

• Volume and Diversity: Compared against other construction datasets, SODA encompasses a broader range of objects and offers comprehensive coverage of construction domain elements for the first time.
• Experimental Evaluation: Two prevalent object detection algorithms, YOLO v3 and YOLO v4, were employed to evaluate the dataset's efficacy. YOLO v3 achieved a mean average precision (mAP) of 71.22% while YOLO v4 reached 81.47%, showcasing the dataset's capability to support typical deep learning models, providing a new benchmark within this domain.

Contributions and Future Work

The SODA dataset represents a significant contribution to computer vision applications in construction, serving as a foundation for training and evaluating deep learning models. It offers:

• Extensive Image Data: Largest collection of construction-related images with diverse classes to date.
• Practical Benchmark: Establishes performance benchmarks for future algorithmic development.
• Open-Source Distribution: The dataset is freely accessible to researchers, facilitating advancements in construction site automation.

In future iterations, the authors aim to enrich SODA by incorporating more categories and expanding the dataset. They also plan to pursue pixel-level annotations and explore automated data collection and annotation techniques to further enhance dataset quality and scale.

The creation of SODA represents a pivotal step towards leveraging deep learning for more efficient and automated construction site management, addressing longstanding challenges in the application of AI in this industry.