Recent advances in camera architectures and associated mathematical representations now enable compressive acquisition of images and videos at low data-rates. While most computer vision applications of today are composed of conventional cameras, which collect a large amount redundant data and power hungry embedded systems, which compress the collected data for further processing, compressive cameras offer the advantage of direct acquisition of data in compressed domain and hence readily promise to find applicability in computer vision, particularly in environments hampered by limited communication bandwidths. However, despite the significant progress in theory and methods of compressive sensing, little headway has been made in developing systems for such applications by exploiting the merits of compressive sensing. In such a setting, we consider the problem of activity recognition, which is an important inference problem in many security and surveillance applications. Since all successful activity recognition systems involve detection of human, followed by recognition, a potential fully functioning system motivated by compressive camera would involve the tracking of human, which requires the reconstruction of atleast the initial few frames to detect the human. Once the human is tracked, the recognition part of the system requires only the features to be extracted from the tracked sequences, which can be the reconstructed images or the compressed measurements of such sequences. However, it is desirable in resource constrained environments that these features be extracted from the compressive measurements without reconstruction. Motivated by this, in this thesis, we propose a framework for understanding activities as a non-linear dynamical system, and propose a robust, generalizable feature that can be extracted directly from the compressed measurements without reconstructing the original video frames. The proposed feature is termed recurrence texture and is motivated from recurrence analysis of non-linear dynamical systems. We show that it is possible to obtain discriminative features directly from the compressed stream and show its utility in recognition of activities at very low data rates.

Diabetic retinopathy (DR) is a common cause of blindness occurring due to prolonged presence of diabetes. The risk of developing DR or having the disease progress is increasing over time. Despite advances in diabetes care over the years, DR remains a vision-threatening complication and one of the leading causes of blindness among American adults. Recent studies have shown that diagnosis based on digital retinal imaging has potential benefits over traditional face-to-face evaluation. Yet there is a dearth of computer-based systems that can match the level of performance achieved by ophthalmologists. This thesis takes a fresh perspective in developing a computer-based system aimed at improving diagnosis of DR images. These images are categorized into three classes according to their severity level. The proposed approach explores effective methods to classify new images and retrieve clinically-relevant images from a database with prior diagnosis information associated with them. Retrieval provides a novel way to utilize the vast knowledge in the archives of previously-diagnosed DR images and thereby improve a clinician's performance while classification can safely reduce the burden on DR screening programs and possibly achieve higher detection accuracy than human experts. To solve the three-class retrieval and classification problem, the approach uses a multi-class multiple-instance medical image retrieval framework that makes use of spectrally tuned color correlogram and steerable Gaussian filter response features. The results show better retrieval and classification performances than prior-art methods and are also observed to be of clinical and visual relevance.

Continuous monitoring of sensor data from smart phones to identify human activities and gestures, puts a heavy load on the smart phone's power consumption. In this research study, the non-Euclidean geometry of the rich sensor data obtained from the user's smart phone is utilized to perform compressive analysis and efficient classification of human activities by employing machine learning techniques. We are interested in the generalization of classical tools for signal approximation to newer spaces, such as rotation data, which is best studied in a non-Euclidean setting, and its application to activity analysis. Attributing to the non-linear nature of the rotation data space, which involve a heavy overload on the smart phone's processor and memory as opposed to feature extraction on the Euclidean space, indexing and compaction of the acquired sensor data is performed prior to feature extraction, to reduce CPU overhead and thereby increase the lifetime of the battery with a little loss in recognition accuracy of the activities. The sensor data represented as unit quaternions, is a more intrinsic representation of the orientation of smart phone compared to Euler angles (which suffers from Gimbal lock problem) or the computationally intensive rotation matrices. Classification algorithms are employed to classify these manifold sequences in the non-Euclidean space. By performing customized indexing (using K-means algorithm) of the evolved manifold sequences before feature extraction, considerable energy savings is achieved in terms of smart phone's battery life.

The increased risk of falling and the worse ability to perform other daily physical activities in the elderly cause concern about monitoring and correcting basic everyday movement. In this thesis, a Kinect-based system was designed to assess one of the most important factors in balance control of human body when doing Sit-to-Stand (STS) movement: the postural symmetry in mediolateral direction. A symmetry score, calculated by the data obtained from a Kinect RGB-D camera, was proposed to reflect the mediolateral postural symmetry degree and was used to drive a real-time audio feedback designed in MAX/MSP to help users adjust themselves to perform their movement in a more symmetrical way during STS. The symmetry score was verified by calculating the Spearman correlation coefficient with the data obtained from Inertial Measurement Unit (IMU) sensor and got an average value at 0.732. Five healthy adults, four males and one female, with normal balance abilities and with no musculoskeletal disorders, were selected to participate in the experiment and the results showed that the low-cost Kinect-based system has the potential to train users to perform a more symmetrical movement in mediolateral direction during STS movement.

There has been tremendous technological advancement in the past two decades. Faster computers and improved sensing devices have broadened the research scope in computer vision. With these developments, the task of assessing the quality of human actions, is considered an important problem that needs to be tackled. Movement quality assessment finds wide range of application in motor control, health-care, rehabilitation and physical therapy. Home-based interactive physical therapy requires the ability to monitor, inform and assess the quality of everyday movements. Obtaining labeled data from trained therapists/experts is the main limitation, since it is both expensive and time consuming.

Motivated by recent studies in motor control and therapy, in this thesis an existing computational framework is used to assess balance impairment and disease severity in people suffering from Parkinson's disease. The framework uses high-dimensional shape descriptors of the reconstructed phase space, of the subjects' center of pressure (CoP) tracings while performing dynamical postural shifts. The performance of the framework is evaluated using a dataset collected from 43 healthy and 17 Parkinson's disease impaired subjects, and outperforms other methods, such as dynamical shift indices and use of chaotic invariants, in assessment of balance impairment.

In this thesis, an unsupervised method is also proposed that measures movement quality assessment of simple actions like sit-to-stand and dynamic posture shifts by modeling the deviation of a given movement from an ideal movement path in the configuration space, i.e. the quality of movement is directly related to similarity to the ideal trajectory, between the start and end pose. The S^1xS^1 configuration space was used to model the interaction of two joint angles in sit-to-stand actions, and the R^2 space was used to model the subject's CoP while performing dynamic posture shifts for application in movement quality estimation.

Modern systems that measure dynamical phenomena often have limitations as to how many sensors can operate at any given time step. This thesis considers a sensor scheduling problem in which the source of a diffusive phenomenon is to be localized using single point measurements of its concentration. With a linear diffusion model, and in the absence of noise, classical observability theory describes whether or not the system's initial state can be deduced from a given set of linear measurements. However, it does not describe to what degree the system is observable. Different metrics of observability have been proposed in literature to address this issue. Many of these methods are based on choosing optimal or sub-optimal sensor schedules from a predetermined collection of possibilities. This thesis proposes two greedy algorithms for a one-dimensional and two-dimensional discrete diffusion processes. The first algorithm considers a deterministic linear dynamical system and deterministic linear measurements. The second algorithm considers noise on the measurements and is compared to a Kalman filter scheduling method described in published work.

Light field imaging is limited in its computational processing demands of high

sampling for both spatial and angular dimensions. Single-shot light field cameras

sacrifice spatial resolution to sample angular viewpoints, typically by multiplexing

incoming rays onto a 2D sensor array. While this resolution can be recovered using

compressive sensing, these iterative solutions are slow in processing a light field. We

present a deep learning approach using a new, two branch network architecture,

consisting jointly of an autoencoder and a 4D CNN, to recover a high resolution

4D light field from a single coded 2D image. This network decreases reconstruction

time significantly while achieving average PSNR values of 26-32 dB on a variety of

light fields. In particular, reconstruction time is decreased from 35 minutes to 6.7

minutes as compared to the dictionary method for equivalent visual quality. These

reconstructions are performed at small sampling/compression ratios as low as 8%,

allowing for cheaper coded light field cameras. We test our network reconstructions

on synthetic light fields, simulated coded measurements of real light fields captured

from a Lytro Illum camera, and real coded images from a custom CMOS diffractive

light field camera. The combination of compressive light field capture with deep

learning allows the potential for real-time light field video acquisition systems in the

future.

Non-line-of-sight (NLOS) imaging of objects not visible to either the camera or illumina-

tion source is a challenging task with vital applications including surveillance and robotics.

Recent NLOS reconstruction advances have been achieved using time-resolved measure-

ments. Acquiring these time-resolved measurements requires expensive and specialized

detectors and laser sources. In work proposes a data-driven approach for NLOS 3D local-

ization requiring only a conventional camera and projector. The localisation is performed

using a voxelisation and a regression problem. Accuracy of greater than 90% is achieved

in localizing a NLOS object to a 5cm × 5cm × 5cm volume in real data. By adopting

the regression approach an object of width 10cm to localised to approximately 1.5cm. To

generalize to line-of-sight (LOS) scenes with non-planar surfaces, an adaptive lighting al-

gorithm is adopted. This algorithm, based on radiosity, identifies and illuminates scene

patches in the LOS which most contribute to the NLOS light paths, and can factor in sys-

tem power constraints. Improvements ranging from 6%-15% in accuracy with a non-planar

LOS wall using adaptive lighting is reported, demonstrating the advantage of combining

the physics of light transport with active illumination for data-driven NLOS imaging.

Several music players have evolved in multi-dimensional and surround sound systems. The audio players are implemented as software applications for different audio hardware systems. Digital formats and wireless networks allow for audio content to be readily accessible on smart networked devices. Therefore, different audio output platforms ranging from multispeaker high-end surround systems to single unit Bluetooth speakers have been developed. A large body of research has been carried out in audio processing, beamforming, sound fields etc. and new formats are developed to create realistic audio experiences.

An emerging trend is seen towards high definition AV systems, virtual reality gears as well as gaming applications with multidimensional audio. Next generation media technology is concentrating around Virtual reality experience and devices. It has applications not only in gaming but all other fields including medical, entertainment, engineering, and education. All such systems also require realistic audio corresponding with the visuals.

In the project presented in this thesis, a new portable audio hardware system is designed and developed along with a dedicated mobile android application to render immersive surround sound experiences with real-time audio effects. The tablet and mobile phone allow the user to control or “play” with sound directionality and implement various audio effects including sound rotation, spatialization, and other immersive experiences. The thesis describes the hardware and software design, provides the theory of the sound effects, and presents demonstrations of the sound application that was created.

Generating real-world content for VR is challenging in terms of capturing and processing at high resolution and high frame-rates. The content needs to represent a truly immersive experience, where the user can look around in 360-degree view and perceive the depth of the scene. The existing solutions only capture and offload the compute load to the server. But offloading large amounts of raw camera feeds takes longer latencies and poses difficulties for real-time applications. By capturing and computing on the edge, we can closely integrate the systems and optimize for low latency. However, moving the traditional stitching algorithms to battery constrained device needs at least three orders of magnitude reduction in power. We believe that close integration of capture and compute stages will lead to reduced overall system power.

We approach the problem by building a hardware prototype and characterize the end-to-end system bottlenecks of power and performance. The prototype has 6 IMX274 cameras and uses Nvidia Jetson TX2 development board for capture and computation. We found that capturing is bottlenecked by sensor power and data-rates across interfaces, whereas compute is limited by the total number of computations per frame. Our characterization shows that redundant capture and redundant computations lead to high power, huge memory footprint, and high latency. The existing systems lack hardware-software co-design aspects, leading to excessive data transfers across the interfaces and expensive computations within the individual subsystems. Finally, we propose mechanisms to optimize the system for low power and low latency. We emphasize the importance of co-design of different subsystems to reduce and reuse the data. For example, reusing the motion vectors of the ISP stage reduces the memory footprint of the stereo correspondence stage. Our estimates show that pipelining and parallelization on custom FPGA can achieve real time stitching.