Short Video wastage.
Central Washington University | 2025
This project develops a sensor-based system for detecting and classifying human motion in indoor environments, with a focus on epilepsy-related pathology. By integrating IMU sensors (accelerometer and gyroscope) and ultrasonic sensing, the system recognizes movements such as walking, sitting, falling, or tremors. The goal is to enable real-time health monitoring for elderly individuals or patients prone to epileptic seizures. The project focuses on the development of systems for sensor interfacing, data acquisition, and signal processing, utilizing techniques such as FFT and time-domain feature extraction to classify human motion. It incorporates machine learning models, including CNNs and RNNs, to identify and differentiate activity patterns with high accuracy. The framework is designed to be robust in noisy indoor environments and suitable for deployment in practical smart home applications.
Short Video wastage.
University of Notre Dame | 2023
This project presents KBL (Key Buffer Logic), a client-side framework designed to improve QoE and network efficiency for short-form video platforms in dynamic mobile environments. KBL integrates Buffer Boundary Management, Pre-loading Strategy, and Bitrate Selection, guided by a real-time throughput estimator for stable network prediction. An extended QoE-efficiency metric is introduced, combining video quality, startup delay, rebuffering, quality switching, residual buffer (L), and data usage (W), calibrated through sensitivity analysis. Experiments across varied network conditions show up to 58% data waste reduction over existing strategies, with QoE preserved or improved. KBL offers a scalable, behavior-aware solution for optimizing mobile video delivery and is well-suited for future deployment in edge-based and smart buffering systems.
Measurement Study across Short Video Services
University of Notre Dame & AT&T Labs Research | 2021 - 2022
Short videos have emerged recently as a popular form of short-duration User Generated Content (UGC) within modern social media. This work provided a comprehensive comparison of four popular short video services. In particular, the work focuses on exploring content characteristics and evaluating the video quality across resolutions for each service. By conducting an experimental study, the work investigates data consumption and finally evaluates achieved QoE under different network scenarios and application configurations. Read more
Congestion Prediction in WiFi Video Streaming.
University of Notre Dame | 2019 - 2020
This research aims to detect the network changes to deliver a better rate adaptation mechanism for adaptive video streaming. This research proposes an algorithm, Congestion Prediction-DASH (CP- DASH), which is designed to prevent stalling and maximize video quality. The CP-DASH algorithm leverages the aggregate MAC protocol data unit information provided by Block Acknowledgement aggregate control packets to allow a WiFi device to passively predict link congestion. Read more
Jumping Motion Recognition.
Figur8. Inc | Feb - Jul 2018
The purpose of this research is to extract the motion features, such as jumping up and down. The data comes from a specific muscle sensor, which records muscle stretch during jumping. Through Recurrent Neural Network modeling, the movement characteristics of the jump are separated, and extracted the jumping features to recognize the specific activities. Read more
Activities Recognition in Smart home.
Ball State Univeristy | 2017
The research proposed a signals solution, Activities Recognition based on Signal Filter (ARSF), to recognize human activities through commodity off-the-shelf radar sensors. The radar sensors detect abnormal and instantaneous motion situations by high-frequency radio signal. The research purpose is to recognize the specific activities of residents in smart homes. During this project, I designed a motion separation algorithm, which extracts the corresponding motion data from the whole data background. Developed an activity recognition system based on distinct features, which distinguishes the distinct motions. Read more
User Localization based on Ultrasonic Beamforming.
Ball State Univeristy | 2016 - 2017
This research work designs and develops a noninvasive distance-based user localization and tracking solution, DiLT, for smart systems. DiLT consists of a mechanical ultrasonic beam-forming design for omni-space sensing, a contrastive divergence learning to localize a user and a binary back-off algorithm to track the motion of the user. During this project, I designed an ultrasonic scanning platform to track the users' location. Developed a location tracking algorithm to filter the environment and noise to identify the specific users' location. Read more