This study evaluates the performance of remote photoplethysmography (rPPG)-based oxygen saturation (SpO2) estimation across different face areas. The information in the red and green channels is extracted through spatial averaging on the detected area, followed by detrending, min-max normalization, and low-pass filtering are performed. Finally, SpO2 is estimated using the Ratio of Ratios method based on Beer-Lambert law. Using the PURE dataset, SpO2 estimation accuracy was evaluated with mean absolute error (MAE). Experimental results showed the entire face provided the highest accuracy, followed by the forehead, left cheek, right cheek, and lips.

This paper proposes a robust algorithm for heart rate estimation using remote photoplethysmography (rPPG). The algorithm employs a combination of adaptive filtering and frequency tracking to enhance the signal-to-noise ratio (SNR) and accurately estimate heart rates from facial videos. The LGI dataset, comprising videos of six participants performing various activities (resting, rotation, talk, gym), was utilized for evaluation. The ground truth heart rate was obtained using a CMS50E pulse oximeter, and a 10-second data window with FFT-based frequency analysis was applied to derive reference heart rates. The proposed method detects the face using Mediapipe API, selects the forehead region of interest (ROI), and extracts RGB signals. The signals undergo preprocessing, motion noise removal via adaptive filtering, and heart rate estimation using an adaptive notch filter. Experimental results demonstrate that the proposed algorithm outperforms existing methods, especially in challenging conditions such as during gym and talk activities.

This research addresses the problem of robot localization in corridor environments using LiDAR (Light Detection and Ranging). Due to the rank deficiency problem in scan matching with LiDAR alone, the accuracy of robot localization may degenerate seriously. This paper proposes an adaptive sampling-based particle filtering method using depth sensors to overcome the rank deficiency problem. The increase in the sample size in particle filters can be considered to solve the problem. But, it may cause much computation cost. In the proposed method, the sample size of the particle set in the proposed method is adjusted adaptively to the confidence of depth sensor data. The performance of the proposed method was test by real experiments in various environments. The experimental results showed that the proposed method was capable of reducing the estimation errors and more accurate than the conventional method.

Visual-inertial odometry (VIO) is a method that leverages sensor data from a camera and an inertial measurement unit (IMU) for state estimation. Whereas conventional VIO has limited capability to estimate scale of translation, the performance of recent approaches has been improved by utilizing depth maps obtained from RGB-D camera, especially in indoor environments. However, the depth map obtained from the RGB-D camera tends to rapidly lose accuracy as the distance increases, and therefore, it is required to develop alternative method to improve the VIO performance in wide environments. In this paper, we argue that leveraging depth map estimated from a deep neural network has benefits to state estimation. To improve the reliability of depth information utilized in VIO algorithm, we propose a kernel-based sampling strategy to filter out depth values with low confidence. The proposed method aims to improve the robustness and accuracy of VIO algorithms by selectively utilizing reliable values of estimated depth maps. Experiments were conducted on real-world custom dataset acquired from underground parking lot environments. Experimental results demonstrate that the proposed method is effective to improve the performance of VIO, exhibiting potential for the use of depth estimation network for state estimation.

In this paper, we propose an improved technique called HAB (Huffman encoding Aware Bitpacking) that enhances the Sprintz method, which is a representative time series data compression technique. The proposed technique boosts compression rates by incorporating Huffman encoding-aware bitpacking in the secondary compression stage. Additionally, it can be applied to terminal nodes or edge nodes with limited available resources, as it does not require separate parameters or storage space for compression. The proposed technique is a lossless method and is suitable for fields that require the generation of artificial intelligence models and accurate data analysis. In the experimental evaluation, the proposed HAB showed an average improvement of 14.7% compared to the existing technique in terms of compression rate.

The automotive industry is transitioning from traditional internal combustion engines to systems powered by motors, batteries, and various electronic control units. Central to this shift is the micro-controller unit, which processes data from various sensors for real-time environmental awareness and control. This paper explores using non-contact magnetic sensors for sensing vehicle inclination as part of a digital twin implementation. Unlike optical or contact sensors, non-contact magnetic sensors offer robust performance in challenging environments, providing consistent and reliable data under varying conditions. To optimize real-time data processing, we propose a double buffer structure to enhance digital signal processing performance in embedded systems. Experiments using a custom sensor-integrated board demonstrate that the double buffer structure with direct memory access-enabled serial peripheral interface significantly reduces data processing time and improves noise reduction filtering. Our results show that the proposed system can greatly enhance the reliability and accuracy of sensor data, crucial for real-time vehicle control systems. In particular, by using the double buffer structure proposed in this paper, it was possible to secure 8.27 times more data compared to raw data, despite performing additional filtering. The techniques outlined have potential applications in various fields, offering enhanced monitoring and optimization capabilities, thus paving the way for more advanced and efficient vehicle control technologies.