Wearable sensors can be used to monitor opioid use and other key behaviors of interest, and to prompt interventions that promote behavioral change. The effectiveness of such systems is threatened by the potential of a subject's deliberate non-compliance (DNC) to the monitoring. We define deliberate non-compliance as the process of giving one's device to someone else when surveillance is on-going. The principal aim of this thesis is to develop an approach to leverage movement and cardiac features from a wearable sensor to detect such deliberate non-compliance by individuals under surveillance for opioid use. Data from 11 participants who presented to the Emergency Department following an opioid overdose was analyzed. Using a personalized machine learning classifier (model), we evaluated if a snippet of blood volume pulse (BVP) and accelerometer data received is coming from the expected participant or an alternate person. Analysis of our classier shows the viability of this approach, as we were able to detect DNC (or compliance) with over 90% accuracy within 3 seconds of its occurrence.

Recording and monitoring vital signs is an essential aspect of home-based healthcare. Using contact sensors to record physiological signals can cause discomfort to patients, especially after prolonged use. Hence, remote physiological measurement approaches have attracted considerable attention as they do not require physical contact with the patient’s skin. Several studies proposed techniques to measure Heart Rate (HR) and Heart Rate Variability (HRV) by detecting the Blood Volume Pulse (BVP) from human facial video recordings while the subject is in a resting condition. In this thesis, we focus on the measurement of HR. We adopt an algorithm that uses the Independent Component Analysis (ICA) to separate the source (physiological) signal from noise in the RGB channels of a facial video. We generalize existing methods to support subject movement during video recording. When a subject is moving, the face may be turned away from the camera. We utilize multiple cameras to enable the algorithm to monitor the vital sign continuously, even if the subject leaves the frame or turns away from a subset of the system’s cameras. Furthermore, we improve the accuracy of existing methods by implementing a light equalization scheme to reduce the effect of shadows and unequal facial light on the HR estimation, a machine learning method to select the most accurate channel outputted by the ICA module, and a regression technique to adjust the initial HR estimate. We systematically test our method on eleven subjects using four cameras. The proposed method decreases the RMSE by 27% compared to the state of the art in the rest condition. When the subject is in motion, the proposed method achieves a RMSE of 1.12 bpm using a single camera and RMSE of 0.96 bpm using multiple camera.

Heart Rate Variability (HRV) analysis has been garnering attention from researchers due to its wide range of applications. Medical researchers have always been interested in Heart Rate (HR) and HRV analysis, but nowadays, investigators from variety of other fields are also probing the subject. For instance, variation in HR and HRV is connected to emotional arousal. Therefore, knowledge from the fields of affective computing and psychology, can be employed to devise machines that understand the emotional states of humans. Recent advancements in non-contact HR and HRV measurement techniques will likely further boost interest in emotional estimation through . Such measurement methods involve the extraction of the photoplethysmography (PPG) signal from the human's face through a camera. The latest approaches apply Independent Component Analysis (ICA) on the color channels of video recordings to extract a PPG signal. Other investigated methods rely on Eulerian Video Magnification (EVM) to detect subtle changes in skin color associated with PPG. The effectiveness of the EVM in HR estimation has well been established. However, to the best of our knowledge, EVM has not been successfully employed to extract HRV feature from a video of a human face. In contrast, ICA based methods have been successfully used for HRV analysis. As we demonstrate in this thesis, these two approaches for HRV feature extraction are highly sensitive to noise. Hence, when we evaluated them in indoor settings, we obtained mean absolute error in the range of 0.012 and 28.4. Therefore, in this thesis, we present two approaches to minimize the error rate when estimating physiological measurements from recorded facial videos using a standard camera. In our first approach which is based on the EVM method, we succeeded in extracting HRV measurements but we could not get rid of high frequency noise, which resulted in a high error percentage for the result of the High frequency (HF) component. Our second proposed approach solved this issue by applying ICA on the red, green and blue (RGB) colors channels and we were able to achieve lower error rates and less noisy signal as compared to previous related works. This was done by using a Buterworth filter with the subject's specific HR range as its Cut-Off. The methods were tested with 12 subjects from the DISCOVER lab at the University of Ottawa, using artificial lights as the only source of illumination. This made it a challenge for us because artificial light produces HF signals which can interfere with the PPG signal. The final results show that our proposed ICA based method has a mean absolute error (MAE) of 0.006, 0.005, 0.34, 0.57 and 0.419 for the mean HR, mean RR, LF, HF and LF/HF respectively. This approach also shows that these physiological parameters are highly correlated with the results taken from the electrocardiography (ECG).