Automatic Food Intake Assessment Using Camera Phones

Obesity is becoming an epidemic phenomenon in most developed countries. The fundamental cause of obesity and overweight is an energy imbalance between calories consumed and calories expended. It is essential to monitor everyday food intake for obesity prevention and management. Existing dietary assessment methods usually require manually recording and recall of food types and portions. Accuracy of the results largely relies on many uncertain factors such as user's memory, food knowledge, and portion estimations. As a result, the accuracy is often compromised. Accurate and convenient dietary assessment methods are still blank and needed in both population and research societies. In this thesis, an automatic food intake assessment method using cameras, inertial measurement units (IMUs) on smart phones was developed to help people foster a healthy life style. With this method, users use their smart phones before and after a meal to capture images or videos around the meal. The smart phone will recognize food items and calculate the volume of the food consumed and provide the results to users. The technical objective is to explore the feasibility of image based food recognition and image based volume estimation. This thesis comprises five publications that address four specific goals of this work: (1) to develop a prototype system with existing methods to review the literature methods, find their drawbacks and explore the feasibility to develop novel methods; (2) based on the prototype system, to investigate new food classification methods to improve the recognition accuracy to a field application level; (3) to design indexing methods for large-scale image database to facilitate the development of new food image recognition and retrieval algorithms; (4) to develop novel convenient and accurate food volume estimation methods using only smart phones with cameras and IMUs. A prototype system was implemented to review existing methods. Image feature detector and descriptor were developed and a nearest neighbor classifier were implemented to classify food items. A reedit card marker method was introduced for metric scale 3D reconstruction and volume calculation. To increase recognition accuracy, novel multi-view food recognition algorithms were developed to recognize regular shape food items. To further increase the accuracy and make the algorithm applicable to arbitrary food items, new food features, new classifiers were designed. The efficiency of the algorithm was increased by means of developing novel image indexing method in large-scale image database. Finally, the volume calculation was enhanced through reducing the marker and introducing IMUs. Sensor fusion technique to combine measurements from cameras and IMUs were explored to infer the metric scale of the 3D model as well as reduce noises from these sensors.

Ultraviolet digital imaging of volcanic plumes : implementation and application to magmatic processes at basaltic volcanoes

Magmatic volatiles play a crucial role in volcanism, from magma production at depth to generation of seismic phenomena to control of eruption style. Accordingly, many models of volcano dynamics rely heavily on behavior of such volatiles. Yet measurements of emission rates of volcanic gases have historically been limited, which has restricted model verification to processes on the order of days or longer. UV cameras are a recent advancement in the field of remote sensing of volcanic SO2 emissions. They offer enhanced temporal and spatial resolution over previous measurement techniques, but need development before they can be widely adopted and achieve the promise of integration with other geophysical datasets. Large datasets require a means by which to quickly and efficiently use imagery to calculate emission rates. We present a suite of programs designed to semi-automatically determine emission rates of SO2 from series of UV images. Extraction of high temporal resolution SO2 emission rates via this software facilitates comparison of gas data to geophysical data for the purposes of evaluating models of volcanic activity and has already proven useful at several volcanoes. Integrated UV camera and seismic measurements recorded in January 2009 at Fuego volcano, Guatemala, provide new insight into the system’s shallow conduit processes. High temporal resolution SO2 data reveal patterns of SO2 emission rate relative to explosions and seismic tremor that indicate tremor and degassing share a common source process. Progressive decreases in emission rate appear to represent inhibition of gas loss from magma as a result of rheological stiffening in the upper conduit. Measurements of emission rate from two closely-spaced vents, made possible by the high spatial resolution of the camera, help constrain this model. UV camera measurements at Kilauea volcano, Hawaii, in May of 2010 captured two occurrences of lava filling and draining within the summit vent. Accompanying high lava stands were diminished SO2 emission rates, decreased seismic and infrasonic tremor, minor deflation, and slowed lava lake surface velocity. Incorporation of UV camera data into the multi-parameter dataset gives credence to the likelihood of shallow gas accumulation as the cause of such events.

Understanding volcanic processes using UV camera measurements of sulfur dioxide and coincident infrasound and seismicity

The exsolution of volatiles from magma maintains an important control on volcanic eruption styles. The nucleation, growth, and connectivity of bubbles during magma ascent provide the driving force behind eruptions, and the rate, volume, and ease of gas exsolution can affect eruptive activity. Volcanic plumes are the observable consequence of this magmatic degassing, and remote sensing techniques allow us to quantify changes in gas exsolution. However, until recently the methods used to measure volcanic plumes did not have the capability of detecting rapid changes in degassing on the scale of standard geophysical observations. The advent of the UV camera now makes high sample rate gas measurements possible. This type of dataset can then be compared to other volcanic observations to provide an in depth picture of degassing mechanisms in the shallow conduit. The goals of this research are to develop a robust methodology for UV camera field measurements of volcanic plumes, and utilize this data in conjunction with seismoacoustic records to illuminate degassing processes. Field and laboratory experiments were conducted to determine the effects of imaging conditions, vignetting, exposure time, calibration technique, and filter usage on the UV camera sulfur dioxide measurements. Using the best practices determined from these studies, a field campaign was undertaken at Volcán de Pacaya, Guatemala. Coincident plume sulfur dioxide measurements, acoustic recordings, and seismic observations were collected and analyzed jointly. The results provide insight into the small explosive features, variations in degassing rate, and plumbing system of this complex volcanic system. This research provides useful information for determining volcanic hazard at Pacaya, and demonstrates the potential of the UV camera in multiparameter studies.

Plenoptic camera : theory and experimental results

The purpose of this research was to develop a working physical model of the focused plenoptic camera and develop software that can process the measured image intensity, reconstruct this into a full resolution image, and to develop a depth map from its corresponding rendered image. The plenoptic camera is a specialized imaging system designed to acquire spatial, angular, and depth information in a single intensity measurement. This camera can also computationally refocus an image by adjusting the patch size used to reconstruct the image. The published methods have been vague and conflicting, so the motivation behind this research is to decipher the work that has been done in order to develop a working proof-of-concept model. This thesis outlines the theory behind the plenoptic camera operation and shows how the measured intensity from the image sensor can be turned into a full resolution rendered image with its corresponding depth map. The depth map can be created by a cross-correlation of adjacent sub-images created by the microlenslet array (MLA.) The full resolution image reconstruction can be done by taking a patch from each MLA sub-image and piecing them together like a puzzle. The patch size determines what object plane will be in-focus. This thesis also goes through a very rigorous explanation of the design constraints involved with building a plenoptic camera. Plenoptic camera data from Adobe © was used to help with the development of the algorithms written to create a rendered image and its depth map. Finally, using the algorithms developed from these tests and the knowledge for developing the plenoptic camera, a working experimental system was built, which successfully generated a rendered image and its corresponding depth map.

VISUAL-INERTIAL INTEGRATION FOR HUMAN MOTION TRACKING AND NAVIGATION IN FREE-LIVING ENVIRONMENTS

The aging population has become a burning issue for all modern societies around the world recently. There are two important issues existing now to be solved. One is how to continuously monitor the movements of those people having suffered a stroke in natural living environment for providing more valuable feedback to guide clinical interventions. The other one is how to guide those old people effectively when they are at home or inside other buildings and to make their life easier and convenient. Therefore, human motion tracking and navigation have been active research fields with the increasing number of elderly people. However, motion capture has been extremely challenging to go beyond laboratory environments and obtain accurate measurements of human physical activity especially in free-living environments, and navigation in free-living environments also poses some problems such as the denied GPS signal and the moving objects commonly presented in free-living environments. This thesis seeks to develop new technologies to enable accurate motion tracking and positioning in free-living environments. This thesis comprises three specific goals using our developed IMU board and the camera from the imaging source company: (1) to develop a robust and real-time orientation algorithm using only the measurements from IMU; (2) to develop a robust distance estimation in static free-living environments to estimate people’s position and navigate people in static free-living environments and simultaneously the scale ambiguity problem, usually appearing in the monocular camera tracking, is solved by integrating the data from the visual and inertial sensors; (3) in case of moving objects viewed by the camera existing in free-living environments, to firstly design a robust scene segmentation algorithm and then respectively estimate the motion of the vIMU system and moving objects. To achieve real-time orientation tracking, an Adaptive-Gain Orientation Filter (AGOF) is proposed in this thesis based on the basic theory of deterministic approach and frequency-based approach using only measurements from the newly developed MARG (Magnet, Angular Rate, and Gravity) sensors. To further obtain robust positioning, an adaptive frame-rate vision-aided IMU system is proposed to develop and implement fast vIMU ego-motion estimation algorithms, where the orientation is estimated in real time from MARG sensors in the first step and then used to estimate the position based on the data from visual and inertial sensors. In case of the moving objects viewed by the camera existing in free-living environments, a robust scene segmentation algorithm is firstly proposed to obtain position estimation and simultaneously the 3D motion of moving objects. Finally, corresponding simulations and experiments have been carried out.

ANKLE IMPEDANCE AND ANKLE ANGLES DURING STEP TURN AND STRAIGHT WALK: IMPLICATIONS FOR THE DESIGN OF A STEERABLE ANKLE-FOOT PROSTHETIC ROBOT

During locomotion, turning is a common and recurring event which is largely neglected in the current state-of-the-art ankle-foot prostheses, forcing amputees to use different steering mechanisms for turning, compared to non-amputees. A better understanding of the complexities surrounding lower limb prostheses will lead to increased health and well-being of amputees. The aim of this research is to develop a steerable ankle-foot prosthesis that mimics the human ankle mechanical properties. Experiments were developed to estimate the mechanical impedance of the ankle and the ankles angles during straight walk and step turn. Next, this information was used in the design of a prototype, powered steerable ankle-foot prosthesis with two controllable degrees of freedom. One of the possible approaches in design of the prosthetic robots is to use the human joints’ parameters, especially their impedance. A series of experiments were conducted to estimate the stochastic mechanical impedance of the human ankle when muscles were fully relaxed and co-contracting antagonistically. A rehabilitation robot for the ankle, Anklebot, was employed to provide torque perturbations to the ankle. The experiments were performed in two different configurations, one with relaxed muscles, and one with 10% of maximum voluntary contraction (MVC). Surface electromyography (sEMG) was used to monitor muscle activation levels and these sEMG signals were displayed to subjects who attempted to maintain them constant. Time histories of ankle torques and angles in the lateral/medial (LM) directions, inversion-eversion (IE), and dorsiflexionplantarflexion (DP) were recorded. Linear time-invariant transfer functions between the measured torques and angles were estimated providing an estimate of ankle mechanical impedance. High coherence was observed over a frequency range up to 30 Hz. The main effect of muscle activation was to increase the magnitude of ankle mechanical impedance in all degrees of freedom of the ankle. Another experiment compared the three-dimensional angles of the ankle during step turn and straight walking. These angles were measured to be used for developing the control strategy of the ankle-foot prosthesis. An infrared camera system was used to track the trajectories and angles of the foot and leg. The combined phases of heel strike and loading response, mid stance, and terminal stance and pre-swing were determined and used to measure the average angles at each combined phase. The Range of motion (ROM) in IE increased during turning while ML rotation decreased and DP changed the least. During the turning step, ankle displacement in DP started with similar angles to straight walk and progressively showed less plantarflexion. In IE, the ankle showed increased inversion leaning the body toward the inside of the turn. ML rotation initiated with an increased medial rotation during the step turn relative to the straight walk transitioning to increased lateral rotation at the toe off. A prototype ankle-foot prosthesis capable of controlling both DP and IE using a cable driven mechanism was developed and assessed as part of a feasibility study. The design is capable of reproducing the angles required for straight walk and step turn; generates 712N of lifting force in plantarflexion, and shows passive stiffness comparable to a nonload bearing ankle impedance. To evaluate the performance of the ankle-foot prosthesis, a circular treadmill was developed to mimic human gait during steering. Preliminary results show that the device can appropriately simulate human gait with loading and unloading the ankle joint during the gait in circular paths.

Driver Safety in Far-side and Far-oblique Crashes: A Study of Patrol Vehicles in the United States of America

The dissertation titled "Driver Safety in Far-side and Far-oblique Crashes" presents a novel approach to assessing vehicle cockpit safety by integrating Human Factors and Applied Mechanics. The methodology of this approach is aimed at improving safety in compact mobile workspaces such as patrol vehicle cockpits. A statistical analysis performed using Michigan state's traffic crash data to assess various contributing factors that affect the risk of severe driver injuries showed that the risk was greater for unrestrained drivers (OR=3.38, p<0.0001) and for incidents involving front and far-side crashes without seatbelts (OR=8.0 and 23.0 respectively, p<0.005). Statistics also showed that near-side and far-side crashes pose similar threat to driver injury severity. A Human Factor survey was conducted to assess various Human-Machine/Human-Computer Interaction aspects in patrol vehicle cockpits. Results showed that tasks requiring manual operation, especially the usage of laptop, would require more attention and potentially cause more distraction. A vehicle survey conducted to evaluate ergonomics-related issues revealed that some of the equipment was in airbag deployment zones. In addition, experiments were conducted to assess the effects on driver distraction caused by changing the position of in-car accessories. A driving simulator study was conducted to mimic HMI/HCI in a patrol vehicle cockpit (20 subjects, average driving experience = 5.35 years, s.d. = 1.8). It was found that the mounting locations of manual tasks did not result in a significant change in response times. Visual displays resulted in response times less than 1.5sec. It can also be concluded that the manual task was equally distracting regardless of mounting positions (average response time was 15 secs). Average speeds and lane deviations did not show any significant results. Data from 13 full-scale sled tests conducted to simulate far-side impacts at 70 PDOF and 40 PDOF was used to analyze head injuries and HIC/AIS values. It was found that accelerations generated by the vehicle deceleration alone were high enough to cause AIS 3 - AIS 6 injuries. Pretensioners could mitigated injuries only in 40 PDOF (oblique) impacts but are useless in 70 PDOF impacts. Seat belts were ineffective in protecting the driver's head from injuries. Head would come in contact with the laptop during a far-oblique (40 PDOF) crash and far-side door for an angle-type crash (70 PDOF). Finite Element analysis head-laptop impact interaction showed that the contact velocity was the most crucial factor in causing a severe (and potentially fatal) head injury. Results indicate that no equipment may be mounted in driver trajectory envelopes. A very narrow band of space is left in patrol vehicles for installation of manual-task equipment to be both safe and ergonomic. In case of a contact, the material stiffness and damping properties play a very significant role in determining the injury outcome. Future work may be done on improving the interiors' material properties to better absorb and dissipate kinetic energy of the head. The design of seat belts and pretensioners may also be seen as an essential aspect to be further improved.