Search

With the recent advancements in Internet of Things (IoT) technology, various fields such as manufacturing, environment, and health care generate diverse types of time series data. This data is collected in real-time and is used for anomaly detection and prediction. However, current time series data analysis does not achieve efficient results due to the use of different preprocessing criteria and different types of deep learning algorithms depending on the purpose. This study addresses these limitations by implementing an automated, customized pipeline called the Time-series Anomaly detection Pipeline (TAP), which efficiently performs prediction and anomaly detection. TAP establishes an integrated deep learning pipeline that allows users to preprocess and model time series data according to their specific environments, enabling accurate prediction and anomaly detection. This approach reduces on-site analysis time and improves the accuracy of predictions and anomaly detection tailored to the environment. [ABSTRACT FROM AUTHOR]

Gyroscopes and camera sensors, widely utilized for motion recognition across industries, are celebrated for their low power consumption and user-friendly interaction. Nevertheless, challenges such as accuracy fluctuations due to calibration methods and computational delays in real-time processing persist. In this innovative study, we introduce ConvIMage, a groundbreaking approach that transforms Inertial Measurement Unit (IMU) sensor time-series data into image data through Convolutional Neural Network (CNN) algorithms, enhancing motion estimation capabilities. ConvIMage ingeniously assembles pattern images by converting IMU-derived time-series data into distinctive color channels, empowering a CNN model to discern intricate movement patterns. Experimental assessments underscore ConvIMage's commendable performance, showcasing accuracies ranging from 83.42% to 85.97%, contingent upon the specific type of motion. This auspicious outcome positions ConvIMage as a robust solution for future IoT service development, leveraging motion data with enhanced accuracy and real-time processing capabilities for a multitude of applications. [ABSTRACT FROM AUTHOR]

Converting boundary images to time-series makes it feasible to perform boundary matching even on a very large image database, which is very important for interactive and fast matching. In recent research, there has been an attempt to perform fast matching considering partial denoising by converting the boundary image into time series. In this paper, to improve performance, we propose an index-building method considering all possible arbitrary denoising parameters for removing arbitrary partial noises. This is a challenging problem since the partial denoising boundary matching must be considered for all possible denoising parameters. We propose an efficient single index-building algorithm by constructing a minimum bounding rectangle(MBR) according to all possible denoising parameters. The results of extensive experiments conducted show that our index-based matching method improves the search performance up to 46.6 ~ 4023.6 times. [ABSTRACT FROM AUTHOR]