Action recognition in intelligent systems

Peng, Han (2022) Action recognition in intelligent systems. Doctoral thesis, Northern Arizona University.

Text Peng_2022_action_recognition_intelligent_systems.pdf - Published Version Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Action recognition is an emerging topic in artificial intelligence, which aims to automatically detect and recognize actions of human and intelligent nodes (vehicles, UAVs) by processing video recordings and data provided by other sensing platforms. Action recognition has been used in a wide range of applications including but not limited to surveillance systems, health-care, athlete training, human-computer interaction modeling, and autonomous vehicles. Action recognition, based on the utilized information acquisition method, can be categorized into three main areas: video-based, radar-based (node-based), and wearable-sensor-based action recognition. However, the definitions are not exclusive and these methods can be overlapping. Developed algorithms are diverse to meet the requirement and constraints of different applications. An important challenge is developing real-time action recognition, which can prohibitively increase the computational cost of the system. In addition, technical implementation challenges can be different in data acquisition based on the utilized platforms. In video-based action recognition, developing highly accurate algorithms requires accommodating different view angles, illumination conditions, camera motions, and background contrast. In the radar-based target tracking method, as an important variant of action recognition, the intrinsic variations among the motion patterns of different species should be taken into account by developing type-specific or customizable models. In wearable-sensor-based action recognition systems, the sensor position can be sensitive for different actions. This project will mainly focus on these three types of action recognition. The aims of this project are developing efficient and fast algorithms for video-based action recognition in different applications, proposing a unified algorithm for radar-based action recognition, and developing novel architecture methods with higher accuracy for wearable-sensor-based action recognition.

Item Type:	Thesis (Doctoral)
Publisher’s Statement:	© Copyright is held by the author. Digital access to this material is made possible by the Cline Library, Northern Arizona University. Further transmission, reproduction or presentation of protected items is prohibited except with permission of the author.
Additional Information:	Artificial intelligence; Action recognition software; video based; Radar based; Wearable-sensor based
Subjects:	Q Science > QA Mathematics Q Science > QA Mathematics > QA76 Computer software
NAU Depositing Author Academic Status:	Student
Department/Unit:	Graduate College > Theses and Dissertations College of Engineering, Informatics, and Applied Sciences > School of Informatics, Computing, and Cyber Systems
Date Deposited:	14 Jun 2023 17:14
Last Modified:	14 Jun 2023 17:14
URI:	https://openknowledge.nau.edu/id/eprint/6026

Actions (login required)

IR Staff Record View

Downloads