Mahsan Rofouei

A Non-invasive Wearable Neck-Cuff System for Real-Time Sleep Monitoring

Sleep is an important part of our lives which affects many life factors such as memory, learning, metabolism and the immune system. Researchers have found correlations between sleep and several diseases such as Chronic Obstructive Pulmonary disease, Chronic Heart Failure, Alzheimers disease, etc. However, sleep data is mainly recorded and diagnosed in sleep labs or in hospitals for some critical cases with high costs. In this work we develop a non-invasive, wearable neck-cuff system capable of real-time monitoring and visualization of physiological signals. These signals are generated from various sensors housed in a soft neck-worn collar and sent via Bluetooth to a cell phone which stores the data. This data is processed and reported to the user or uploaded to the cloud and/or to a local PC. With this system we are able to monitor peoples sleep continuously in a non-invasive and low cost method while at the same time collect a large database for sleep data which may benefit future advances in new findings and possibly enable a diagnosis of other diseases. We show as one of the applications of our system the possible detection of obstructive sleep apnea which is a common sleep disorder.

Your phone or mine?: fusing body, touch and device sensing for multi-user device-display interaction

Determining who is interacting with a multi-user interactive touch display is challenging. We describe a technique for associating multi-touch interactions to individual users and their accelerometer-equipped mobile devices. Real-time device accelerometer data and depth camera-based body tracking are compared to associate each phone with a particular user, while body tracking and touch contacts positions are compared to associate a touch contact with a specific user. It is then possible to associate touch contacts with devices, allowing for more seamless device-display multi-user interactions. We detail the technique and present a user study to validate and demonstrate a content exchange application using this approach.

Reliability-Aware Optimization for DVS-Enabled Real-Time Embedded Systems

Power and energy consumption has emerged as the premier and most constraining aspect in modern computational systems. Dynamic voltage scheduling (DVS) has been provably one of the most effective techniques used to achieve low power specification. On the other hand, as the feature size of logic gates (and transistors) is becoming smaller and smaller, the effect of soft error rates caused by single event upsets (SEUs) becomes exponentially greater. Lowering supply voltage to save energy increases soft error rates caused by SEU for two reasons: I) lower voltage makes digital circuits more prone to soft errors and II) reduction in supply voltage, increases the duration of process which increases the chances of being hit by SEU. In this paper, we propose an optimal methodology for DVS on a task graph with consideration of soft error rate. We consider the effects of voltage on SEU and incorporate this dependency in our formulation to develop a new method for energy optimization under SEU constraints. We also propose a convex programming formulation that can be solved efficiently and optimally. We show the effectiveness of our optimal results by simulation on TGFF benchmarks.

SmartLDWS: A robust and scalable lane departure warning system for the smartphones

Lane Departure Warning Systems (LDWS) have recently become an integral part of many advance vision-based drive assistance systems. However, high cost and the requirement of professional installation have limited such systems to mostly commercial or luxury vehicles. To help bring the technology to the mainstream market, we have leveraged the popularity of smartphones and built SmartLDWS, the first LDWS that runs on these devices. SmartLDWS employs a novel lane detection algorithm that is both robust and scalable to overcome poor camera quality and limited processing power faced by most smartphones. Experimental results show that the system performs reliably with extremely low false-positive under different weather and lighting conditions, detecting various types of lane markings at over 30fps.

Computing with uncertainty in a smart textile surface for object recognition

A wearable surface capable of performing object recognition on objects placed on it has many applications in health care such as surgery, assisted living posture monitoring, specifically movement of body parts during sleep and etc. The flexibility and wearability of textile material allows its widespread applications in body-worn contexts. In this work, we propose a portable and wearable smart textile surface which is capable of performing object recognition on a set of prior known objects. We integrate data from multiple sensors to gain knowledge about objects in the environment. The uncertainty present in such systems can lead to inaccurate interpretation of the data which is crucial in various medical applications. The most significant part of this uncertainty is due to effects of multiple sensors on each other. We look at different sources of uncertainties in such systems and formulate them. We modify vision algorithm to account for these uncertainties and in the end we present precision bounds for the accuracy of the system.

Fast GPU-based space-time correlation for activity recognition in video sequences

Action recognition is becoming an important component of many computer vision applications such as video surveillance, video indexing and browsing. However most of the space time approaches to action recognition are very computationally expensive which prevents us from using them in real-time applications. This paper describes how Graphic Processing Units (GPUs) can be used in the field of action recognition to speed up this process. We implement a space-time behavior based correlation scheme on NVIDIA Quadro FX 5600 GPU and gain a 50x speedup over its counterpart CPU implementation.

Energy efficient E-textile based portable keyboard

We have created sensor architecture, data collection and processing techniques for an E-Textile wireless keyboard. We leverage the inherent properties of E-Textiles to produce optimized architecture for energy efficient sensing. Novel techniques such as one where each sensor senses several events (activations of different keys) and each event is sensed by three sensors and flexible interleaved sensing and data processing result in up to a factor of 30 energy reduction over the system where each key is sensed by exactly one sensor. We build the keyboard and test it on multiple subjects.

Opportunistic hierarchical classification for power optimization in wearable movement monitoring systems

Patient monitoring systems are becoming increasingly important in accurately diagnosing and treating growing worldwide chronic conditions especially the obesity epidemic. The ubiquitous nature of wearable sensors, such as the readily available embedded accelerometers in smart phones, provides physicians with an opportunity to remotely monitor their patients daily activity. There have been several developments in the area of activity recognition using wearable sensors. However, due to power constraints, resource efficient algorithms are necessary in order to perform accurate realtime activity recognition while consuming minimal energy. In this paper, we present a two-tier architecture for optimizing power consumption in such systems. While the first tier relies on a hierarchical classification approach, the second one manages the activation and deactivation of the classification system. We demonstrate this using a series of binary Support Vector Machine classifiers. The proposed approach, however, is classifier independent. Experimenting with subjects performing different daily activities such as walking, going upstairs and down-stairs, standing and sitting, our approach achieves a power savings of 87%, while maintaining 92% classification accuracy.

Optimization intensive energy harvesting

Instrumented Medical Shoes (MSs) are equipped with a variety of sensors for measurement of quantities such as pressure, acceleration, and temperature which are often greatly beneficial in numerous diagnosis, monitoring, rehabilitation, and other medical tasks. One of primary limiting factors of MSs is their energy sensitivity. In order to overcome this limitation, we have developed an optimization intensive approach for energy harvesting. Our goal is to size and position a single piezoelectric transducer for energy generation in a medical shoe in such a way that maximal energy is collected and/or specified maximal voltage is achieved while collecting energy. We propose a scenario approach that provides statistically sound solution and evaluate our approach using our medical shoe simulator for subject specific energy harvesting and generic MS scavenging. We could get 3.7X energy gain compare to smallest size sensor and 1.3X energy gain compared to sensor with the size of a shoe.

Machine Learning-Based Adaptive Wireless Interval Training Guidance System

Interval training has been shown to improve the physical and psychological performance of users, in terms of fatigue level, cardiovascular build-up, hemoglobin concentration, and self-esteem. Despite the benefits, there is no known automated method for formulating and tailoring an optimized interval training protocol for a specific individual that maximizes the amount of calories burned while limiting fatigue. Additionally, an application that provides the aforementioned optimal training protocol must also provide motivation for repetitious and tedious exercises necessary to improve a patient’s adherence. This paper presents a system that efficiently formulates an optimized interval training method for each individual by using data mining schemes on attributes, conditions, and data gathered from individuals exercise sessions. This system uses accelerometers embedded within iPhones, a Bluetooth pulse oximeter, and the Weka data mining tool to formulate optimized interval training protocols and has been shown to increase the amount of calories burned by 29.54% as compared to the modified Tabata interval training protocol. We also developed a behavioral cueing system that uses music and performance feedback to provide motivation during interval training exercise sessions. By measuring a user’s performance through sensor readings, we are able to play songs that match the user’s workout plan. A hybrid collaborative, content, and context-aware filtering algorithm incorporates the user’s music preferences and the exercise speed to enhance performance.