**1. Introduction**

A better understanding of human behavior benefits individuals on a large scale, including healthcare, well-being, social interaction, life assistance, etc. Thus human activity recognition (HAR) has been tremendously explored in recent years, driven by the enormous technical advances in sensing, computation, and immense human-centric user scenarios. The explosive advancement in machine learning and hardware architecture has dramatically improved the accuracy and robustness of HAR tasks and enabled the technique to be deployed at the far edge near the body. Besides the computational ability, the sensing technique plays a fundamental and critical role in HAR tasks. Therefore, a broader range of sensing modalities has been explored in recent years, aiming to boost the development of reliable body activity digitalized recording. The proposed sensing modalities range from traditional motion sensing methods such as accelerometers, to novel TOF-based sensing such as mmWave, from neural network-aided image processing for activity abstraction to very straightforward proximity detection approaches such as RF-tags.

To acquire a comprehensive overview of the state-of-the-art sensing modalities in human activity recognition, categorization of the adopted sensors is an efficient approach for a deeper understanding of the sensing medium. Researchers have already categorized related sensors into different classes, such as active and passive sensors depending on the need for external excitation [1], or intrusive and non-intrusive sensors depending on the interference of the sensors in the process flow [2,3]. With a further step, we elaborately categorized the HAR-related sensing modalities into five classes depending on the following

**Citation:** Bian, S.; Liu, M.; Zhou, B.; Lukowicz, P. The State-of-the-Art Sensing Techniques in Human Activity Recognition: A Survey. *Sensors* **2022**, *22*, 4596. https:// doi.org/10.3390/s22124596

Academic Editors: Tanja Schultz, Hui Liu and Hugo Gamboa

Received: 13 May 2022 Accepted: 16 June 2022 Published: 17 June 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

sensing principles: kinematic sensing, field-based sensing, wave-based sensing, physiological sensing, and hybrid or other approaches, as Figure 1 presents. We enumerated most of the sensing modalities within each class with an in-depth description of the sensing tricks in the HAR tasks.

**Figure 1.** Sensing techniques in human activity recognition.

#### *1.1. Relevant Surveys*

Despite the enormous scope of sensing modalities in HAR tasks, related survey works are limited. The existing surveys on HAR sensing are primarily focused either on a specific scenario (such as wearable sensing or video-based sensing) or on full-stack presentation of both sensing and data processing techniques, which results in a weak focus on HARrelated sensing techniques. Table 1 lists the latest HAR sensing-related surveys in recent years from the literature. Those high-related surveys (as well as other references listed in this paper) are first searched using keywords such as human activity recognition, survey, overview, and sensing technique, from platforms including Google Scholar, IEEE Xplore, Microsoft Academic, etc. Second, the survey papers cited in the searched surveys were also considered. As can be seen, nearly all the exiting surveys only focused on a specific domain of HAR sensing techniques, such as device-free sensors [4], smartphone sensors [5], radar sensors [6], etc. Such surveys could supply a detailed research result on the particular sensing domain but lack focus on the adopted sensors in HAR. In contrast, there are only a few surveys [7,8] that supply thorough sensor modalities. However, an in-depth introduction and comparison of the sensing tricks is still lacking.

**Table 1.** Surveys on HAR sensing techniques.



**Table 1.** *Cont.*

#### *1.2. Paper Aims and Contribution*

This work tries to fill the gap by presenting an extensive and in-depth survey on the state-of-the-art sensing modalities in HAR tasks, aiming to supply a solid understanding of most sensing modalities for researchers in the community. Overall, we provide the following contributions in this survey:


This survey is constructed as follows: in Section 1, we briefly stated the motivation of this survey considering the existing works and the development of the state-of-the-art HAR sensing techniques. We then summarized and categorized all human activities in the research scope according to the activity attribute in Section 2, followed by a brief description of the general process of the HAR task. Section 3 showed our categorization of the current HAR sensing techniques and gave an in-depth and extensive description of each sensing modality, followed by a summary regarding eight critical sensing performance metrics. Sections 4 and 5 presented a few outlooks into the future development of the HAR-related sensing techniques and the conclusion of our work.
