Human Activity Recognition

Human activity recognition is an AI technology that identifies the behaviors of observed subjects using sensor data collected from devices such as wearable or environmental sensors. It serves as a foundational technology for realizing future smart societies and industries, including smart factories, smart healthcare, and smart homes. At the Real-World Intelligence Laboratory, we focus particularly on recognizing and understanding complex activities in environments such as smart factories and life science experiments. We develop real-world AI technologies that can be applied to coaching systems and industrial robotics. In addition, we construct large-scale activity recognition datasets and develop benchmarks for foundation models in activity recognition.

Related Publications

Self-Supervised Learning for Complex Activity Recognition Through Motif Identification Learning

Qingxin Xia, Jaime Morales, Yongzhi Huang, Takahiro Hara, Kaishun Wu, Hirotomo Oshima, Masamitsu Fukuda, Yasuo Namioka, Takuya Maekawa
IEEE Transactions on Mobile Computing, (May 2025)

In this study, we propose a novel self-supervised learning method for complex activity recognition through motif identification learning.

Self-Supervised LearningHuman Activity RecognitionMotif IdentificationDeep Learning

OpenPack: A Large-Scale Dataset for Recognizing Packaging Works in IoT-Enabled Logistic Environments.

Naoya Yoshimura, Jaime Morales, Takuya Maekawa, Takahiro Hara
IEEE International Conference on Pervasive Computing and Communications, (Mar. 2024)

This paper introduces OpenPack, a large-scale dataset for recognizing packaging work in IoT-enabled logistic environments.

Human Activity RecognitionWearable SensorsLogisticsDataset

WeakCounter: Acceleration-based Repetition Counting of Actions with Weakly Supervised Learning

Yuuki Nishino, Takuya Maekawa, Takahiro Hara
ACM International Symposium on Wearable Computers, (Sep. 2021)

This paper proposes WeakCounter, a method for counting action repetitions from acceleration data using weakly supervised learning.

Weakly Supervised LearningWearable Sensors

Human Activity Recognition with Deep Reinforcement Learning using the Camera of a Mobile Robot

Teerawat Kumrai, Joseph Korpela, Takuya Maekawa, Yen Yu, Ryota Kanai
IEEE International Conference on Pervasive Computing and Communications, (Mar. 2020)

This paper proposes a human activity recognition method using deep reinforcement learning with the camera of a mobile robot.

Human Activity RecognitionDeep Reinforcement LearningMobile Robot

Evaluating Tooth Brushing Performance With Smartphone Sound Data

Joseph Korpela, Ryosuke Miyaji, Takuya Maekawa, Kazunori Nozaki, Hiroo Tamagawa
ACM International Joint Conference on Pervasive and Ubiquitous Computing, (2015)

This paper proposes a method for evaluating tooth brushing performance using smartphone sound data.

Health MonitoringAcoustic SensingHuman Activity Recognition