Year: 2025
Type: article
Abstract: The issue of personal safety for crime prevention has become a significant societal concern. Existing software on smartwatches developed for personal protection might provide GPS location tracking and emergency reporting, but this is limited to proactively detecting and responding to actual at-risk situations. This paper presents a real-time motion detection algorithm for smartwatches that utilizes an accelerometer to identify at-risk movements when a wearer is under threat. Daily activities, including walking, running, desk work, and being threatened, are distinguished by a machine learning-based alarm application. A total of 5534 data points across four classes were collected from experiments. The proposed 1D-CNN model exhibited the highest performance in comparison with SVM, k-NN, random forest, SGD. Additionally, our comparative analysis of using time-domain versus frequency-domain data in machine learning revealed that frequency-domain features offer advantages in both accuracy and real-time performance. Finally, the proposed inference model was implemented as a smartwatch application that can detect at-risk situations in real time. The application was tested in real-world scenarios, showcasing the effectiveness of personal safety.
Source: Applied Sciences
DOI: 10.3390/app15041842
Authors: Haeun Oh, Jaehyun Yoo
Institution: Sungshin Women's University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,896
Funder: Korean National Police Agency
Award: G5942663084
OpenAlex ID: W4407353111