HealCam: Energy-efficient and privacy-preserving human vital cycles monitoring on camera-enabled smart devices

Abstract

In this paper, we present HealCam, an energy-efficient and privacy-preserving human vital signs (e.g., respiration cycles) monitoring system on camera-enabled smart devices. HealCam incorporates the related theories of compressive sensing in its system to reduce the sampling rate while preserving data privacy. HealCam saves significant cost on video processing via low-rate and non-uniform random sampling. It also provides a privacy-preserved data collection and enquiry service via lightweight compressive encryption and decryption scheme. According to our evaluations on real datasets, HealCam achieves high accuracy on respiration cycles reconstruction with extremely low average frame rate, i.e., 1 FPS, via non-uniform random sampling compared with traditional uniform sampling strategy. Then we implement HealCam on smartphones to evaluate its resource consumption. The results show that, HealCam is 23.6 times more energy efficient and 26.7 times faster than the original approach on video processing. Its data encryption component is 172 times faster while consumes only 1.01% energy compared with the corresponding state-of-the-art.

Introduction

Human vital signs, e.g., respiration and heartbeat cycles, are important references for medical diagnoses and healthcare applications. These vital cycles can be monitored by the dedicated equipments in hospital, e.g., an ElectroCardioGram (ECG)  [1] for heartbeat cycles monitoring. However, these equipments are expensive, cumbersome and require special expertise to use; therefore are not appropriate for usage in collecting daily health statistics of patients or the elderly in nursing-house, aged-care facilities or even at home. As the vision of smart healthcare, the future of medical applications should leverage the pervasive smart devices under different environments to provide intelligent, seamless and non-intrusive healthcare services. Therefore, the traditional dedicated and cumbersome equipments are not appropriate.

Extracting the cycles of respiration and heartbeat from continues video recordings of human faces has been extensively studied in the literature [2] and the current state-of-the-art is based on detecting the subtle head motions [3] using the embedded cameras on high performance PCs. The video-based approaches are quite promising as they do not require any special expertise to use and the high-resolution on-board cameras are now pervasive on different embedded smart devices including smartphones, Augmented Reality/Virtual Reality (AR/VR) glasses, tablets and etc. (see Fig. 1). The flexibility and user experience of the video-based approaches will be significantly improved if implemented on the camera-enabled smart devices. However, video processing is known to be computationally prohibitive for resource-constrained smart devices and the computational issues are ignored in previous research.

A straightforward solution for reducing the video computation cost on smart devices is offloading the computationally intensive tasks to the high performance computing facilities, such as local servers or remote cloud computing services [4]. However, offloading the video recordings of individuals’ faces is quite privacy-intrusive. Moreover, the extracted vital cycles themselves are significantly private as it can be used to infer individuals’ health status. Individuals’ health related data has substantially commercial values and it is highly likely that it will be targeted by hackers in the future in the same way that credit card information is now.

In this paper, we focus on monitoring the respiration cycles and addressing the above issues, we propose a novel energy-efficient and privacy-preserving human respiration cycles monitoring system, HealCam, on camera-enabled smart devices by incorporating relevant theories of compressive sensing. The contributions of this paper are as follows,

• In this paper, we propose HealCam for video-based human respiration cycles monitoring on camera-enabled smart devices. To the best of our knowledge, this is the first relevant system running on resource-constrained devices.

• As the minimal frame rate of uniform approach is bounded by the theory of Nyquist frequency and noises, to reduce the sampling rate, we propose a non-uniform random sampling strategy inspired by compressive sensing, instead of traditional uniform sampling to reduce the local video computation cost on the smart devices. According to the evaluation on the real datasets, it achieves significantly higher reconstruction accuracy at extremely low sampling rate compared with the uniform sampling strategy.

• We incorporate a lightweight compressive encryption and decryption algorithm in HealCam, based on compressive sensing, for privacy-preserved data collection and enquiry with smart devices.

• At last, we implement HealCam on smartphones to evaluate its resource consumptions. The results show that HealCam significantly reduces the energy consumption on both local video computing and data encryption.

This paper is organized as followed: Section 2 gives introduction about the work related to monitoring human vital cycles and corresponding encryption methods. Section 3 mentions the detail of HealCam’s system architecture. In Section 4 results of experiments conducted on real world dataset are given. Finally Section 5 concludes the paper.