"BasicAI provided us with an excellent overall experience. We were working on a smart toothbrush project, annotating videos of toothbrushing sessions. One aspect of the platform that really helped us was the ability to directly supervise the annotation process and tailor the labeling to meet our specific needs."

Background

This project, a graduation design at UCLA, aims to develop a prototype smart toothbrush that monitors brushing behavior using Inertial Measurement Unit (IMU) sensors, including accelerometers, gyroscopes, and magnetometers. The project stems from the need for dental disease prevention and the limitations of existing monitoring methods. While good brushing habits are crucial for plaque removal, current assessment techniques are primarily conducted in laboratory environments, failing to accurately reflect daily brushing behaviors.

The core objective of the project is to develop a predictive model to improve brushing efficiency and oral health. To achieve this, the research team collected and analyzed over 1,000 toothbrushing videos. By applying advanced algorithms such as time series analysis and deep learning models, the team was able to evaluate various brushing techniques and habits. This multidimensional data analysis not only provided valuable insights for future smart toothbrush designs but also supported the development of personalized oral care plans. Furthermore, this research advanced the application of Digital Health Interventions (DHI) and health data analytics in the medical field, particularly in oral health management.

Challenge

During the research, the team faced various data annotation challenges that required precise handling and meticulous analysis to ensure data reliability and model accuracy. The main challenges included:

Precise area identification: Accurately determining the specific oral region being brushed (e.g., upper/lower teeth, left/right side). This often required annotators to have professional knowledge or keen observation skills, as it was sometimes difficult to clearly see the exact position of the toothbrush in the video.

Accurate time labeling: The task demanded extremely precise recording of start and end times for each brushing area, down to individual samples. This meant annotators had to carefully watch the videos and combine chart data to accurately judge when the toothbrush started and stopped brushing a particular area.

Handling transition areas: Determining when to end one time period and start the next as the toothbrush moves between different areas. This required annotators to keenly capture the toothbrush's movement and distinguish between transitional phases and stable brushing phases.

Data consistency: The task required annotations for different time periods to be non-overlapping and each period to meet duration requirements (at least 2 seconds). This necessitated annotators to maintain consistency while annotating, avoiding omissions or duplicate markings.

Integration of multiple data sources: The need to reference both video and charts (such as gyroscope and accelerometer data) meant that annotators had to be able to integrate and understand information from multiple data types to accurately mark brushing areas and time periods.

Solution

The BasicAI team provided high-quality data annotation services for the UCLA project, effectively helping them overcome various challenges.

The BasicAI platform has the capability to segment videos into multiple parts and assign specific labels to each part. This feature significantly improved the efficiency of video annotation, especially when dealing with fast-moving dynamic scenes, ensuring that every detail could be precisely annotated. This capability was particularly important when annotating different areas and periods of brushing behavior, as it required a high degree of precision and attention to detail.

Throughout the annotation process, the UCLA team was able to directly supervise and curate, ensuring that the annotation process met their research needs. To support this process, the BasicAI team provided detailed documentation and annotation guidelines and conducted multiple training sessions for the annotators. These training sessions not only covered basic annotation techniques but also delved into the nuances of various brushing behaviors. This comprehensive training enabled annotators to more accurately identify and label brushing areas and their changes, improving the overall quality and consistency of annotations.

Moreover, the BasicAI team implemented multiple rounds of review and quality checks to ensure the accuracy of initial annotations. During this process, any potential errors could be promptly identified and corrected, thus avoiding issues that might affect the final model performance. These multi-layered quality control measures guaranteed the high precision of the final data, providing the UCLA team with reliable data support.

Through these effective solutions, the UCLA team successfully overcame challenges, improved data quality, and laid a solid foundation for the smooth progress of their research.

Result

Through BasicAI’s data labeling service, researchers were able to deeply analyze and understand users' brushing behaviors. This high-precision labeled data provided valuable insights for the smart toothbrush project, allowing the research team to more effectively optimize brushing techniques and behavior recognition algorithms. The model's behavior detection accuracy significantly improved, enabling more precise identification and classification of brushing actions and areas. Researchers also developed personalized brushing recommendations based on user data, helping users improve their brushing habits and enhance oral health. The high-quality labeled data supported the design and development of the smart toothbrush prototype, helping the team optimize the product’s functionality and performance, and significantly enhancing the user experience.

The team expressed satisfaction with the collaboration, and both parties look forward to continuing their partnership in future AI research projects.