Matching Items (19)
Filtering by
- All Subjects: deep learning
- All Subjects: Human activity recognition
- Creators: Turaga, Pavan
- Member of: ASU Electronic Theses and Dissertations
Description
We solve the problem of activity verification in the context of sustainability. Activity verification is the process of proving the user assertions pertaining to a certain activity performed by the user. Our motivation lies in incentivizing the user for engaging in sustainable activities like taking public transport or recycling. Such incentivization schemes require the system to verify the claim made by the user. The system verifies these claims by analyzing the supporting evidence captured by the user while performing the activity. The proliferation of portable smart-phones in the past few years has provided us with a ubiquitous and relatively cheap platform, having multiple sensors like accelerometer, gyroscope, microphone etc. to capture this evidence data in-situ. In this research, we investigate the supervised and semi-supervised learning techniques for activity verification. Both these techniques make use the data set constructed using the evidence submitted by the user. Supervised learning makes use of annotated evidence data to build a function to predict the class labels of the unlabeled data points. The evidence data captured can be either unimodal or multimodal in nature. We use the accelerometer data as evidence for transportation mode verification and image data as evidence for recycling verification. After training the system, we achieve maximum accuracy of 94% when classifying the transport mode and 81% when detecting recycle activity. In the case of recycle verification, we could improve the classification accuracy by asking the user for more evidence. We present some techniques to ask the user for the next best piece of evidence that maximizes the probability of classification. Using these techniques for detecting recycle activity, the accuracy increases to 93%. The major disadvantage of using supervised models is that it requires extensive annotated training data, which expensive to collect. Due to the limited training data, we look at the graph based inductive semi-supervised learning methods to propagate the labels among the unlabeled samples. In the semi-supervised approach, we represent each instance in the data set as a node in the graph. Since it is a complete graph, edges interconnect these nodes, with each edge having some weight representing the similarity between the points. We propagate the labels in this graph, based on the proximity of the data points to the labeled nodes. We estimate the performance of these algorithms by measuring how close the probability distribution of the data after label propagation is to the probability distribution of the ground truth data. Since labeling has a cost associated with it, in this thesis we propose two algorithms that help us in selecting minimum number of labeled points to propagate the labels accurately. Our proposed algorithm achieves a maximum of 73% increase in performance when compared to the baseline algorithm.
ContributorsDesai, Vaishnav (Author) / Sundaram, Hari (Thesis advisor) / Li, Baoxin (Thesis advisor) / Turaga, Pavan (Committee member) / Arizona State University (Publisher)
Created2013
Description
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.
ContributorsSivakumar, Aswin (Author) / Turaga, Pavan (Thesis advisor) / Spanias, Andreas (Committee member) / Papandreou-Suppappola, Antonia (Committee member) / Arizona State University (Publisher)
Created2014
Description
In this thesis, we consider the problem of fast and efficient indexing techniques for time sequences which evolve on manifold-valued spaces. Using manifolds is a convenient way to work with complex features that often do not live in Euclidean spaces. However, computing standard notions of geodesic distance, mean etc. can get very involved due to the underlying non-linearity associated with the space. As a result a complex task such as manifold sequence matching would require very large number of computations making it hard to use in practice. We believe that one can device smart approximation algorithms for several classes of such problems which take into account the geometry of the manifold and maintain the favorable properties of the exact approach. This problem has several applications in areas of human activity discovery and recognition, where several features and representations are naturally studied in a non-Euclidean setting. We propose a novel solution to the problem of indexing manifold-valued sequences by proposing an intrinsic approach to map sequences to a symbolic representation. This is shown to enable the deployment of fast and accurate algorithms for activity recognition, motif discovery, and anomaly detection. Toward this end, we present generalizations of key concepts of piece-wise aggregation and symbolic approximation for the case of non-Euclidean manifolds. Experiments show that one can replace expensive geodesic computations with much faster symbolic computations with little loss of accuracy in activity recognition and discovery applications. The proposed methods are ideally suited for real-time systems and resource constrained scenarios.
ContributorsAnirudh, Rushil (Author) / Turaga, Pavan (Thesis advisor) / Spanias, Andreas (Committee member) / Li, Baoxin (Committee member) / Arizona State University (Publisher)
Created2012
Description
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.
ContributorsKulkarni, Kuldeep Sharad (Author) / Turaga, Pavan (Thesis advisor) / Spanias, Andreas (Committee member) / Frakes, David (Committee member) / Arizona State University (Publisher)
Created2012
Description
Computer Vision as a eld has gone through signicant changes in the last decade.
The eld has seen tremendous success in designing learning systems with hand-crafted
features and in using representation learning to extract better features. In this dissertation
some novel approaches to representation learning and task learning are studied.
Multiple-instance learning which is generalization of supervised learning, is one
example of task learning that is discussed. In particular, a novel non-parametric k-
NN-based multiple-instance learning is proposed, which is shown to outperform other
existing approaches. This solution is applied to a diabetic retinopathy pathology
detection problem eectively.
In cases of representation learning, generality of neural features are investigated
rst. This investigation leads to some critical understanding and results in feature
generality among datasets. The possibility of learning from a mentor network instead
of from labels is then investigated. Distillation of dark knowledge is used to eciently
mentor a small network from a pre-trained large mentor network. These studies help
in understanding representation learning with smaller and compressed networks.
The eld has seen tremendous success in designing learning systems with hand-crafted
features and in using representation learning to extract better features. In this dissertation
some novel approaches to representation learning and task learning are studied.
Multiple-instance learning which is generalization of supervised learning, is one
example of task learning that is discussed. In particular, a novel non-parametric k-
NN-based multiple-instance learning is proposed, which is shown to outperform other
existing approaches. This solution is applied to a diabetic retinopathy pathology
detection problem eectively.
In cases of representation learning, generality of neural features are investigated
rst. This investigation leads to some critical understanding and results in feature
generality among datasets. The possibility of learning from a mentor network instead
of from labels is then investigated. Distillation of dark knowledge is used to eciently
mentor a small network from a pre-trained large mentor network. These studies help
in understanding representation learning with smaller and compressed networks.
ContributorsVenkatesan, Ragav (Author) / Li, Baoxin (Thesis advisor) / Turaga, Pavan (Committee member) / Yang, Yezhou (Committee member) / Davulcu, Hasan (Committee member) / Arizona State University (Publisher)
Created2017
Description
The performance of most of the visual computing tasks depends on the quality of the features extracted from the raw data. Insightful feature representation increases the performance of many learning algorithms by exposing the underlying explanatory factors of the output for the unobserved input. A good representation should also handle anomalies in the data such as missing samples and noisy input caused by the undesired, external factors of variation. It should also reduce the data redundancy. Over the years, many feature extraction processes have been invented to produce good representations of raw images and videos.
The feature extraction processes can be categorized into three groups. The first group contains processes that are hand-crafted for a specific task. Hand-engineering features requires the knowledge of domain experts and manual labor. However, the feature extraction process is interpretable and explainable. Next group contains the latent-feature extraction processes. While the original feature lies in a high-dimensional space, the relevant factors for a task often lie on a lower dimensional manifold. The latent-feature extraction employs hidden variables to expose the underlying data properties that cannot be directly measured from the input. Latent features seek a specific structure such as sparsity or low-rank into the derived representation through sophisticated optimization techniques. The last category is that of deep features. These are obtained by passing raw input data with minimal pre-processing through a deep network. Its parameters are computed by iteratively minimizing a task-based loss.
In this dissertation, I present four pieces of work where I create and learn suitable data representations. The first task employs hand-crafted features to perform clinically-relevant retrieval of diabetic retinopathy images. The second task uses latent features to perform content-adaptive image enhancement. The third task ranks a pair of images based on their aestheticism. The goal of the last task is to capture localized image artifacts in small datasets with patch-level labels. For both these tasks, I propose novel deep architectures and show significant improvement over the previous state-of-art approaches. A suitable combination of feature representations augmented with an appropriate learning approach can increase performance for most visual computing tasks.
The feature extraction processes can be categorized into three groups. The first group contains processes that are hand-crafted for a specific task. Hand-engineering features requires the knowledge of domain experts and manual labor. However, the feature extraction process is interpretable and explainable. Next group contains the latent-feature extraction processes. While the original feature lies in a high-dimensional space, the relevant factors for a task often lie on a lower dimensional manifold. The latent-feature extraction employs hidden variables to expose the underlying data properties that cannot be directly measured from the input. Latent features seek a specific structure such as sparsity or low-rank into the derived representation through sophisticated optimization techniques. The last category is that of deep features. These are obtained by passing raw input data with minimal pre-processing through a deep network. Its parameters are computed by iteratively minimizing a task-based loss.
In this dissertation, I present four pieces of work where I create and learn suitable data representations. The first task employs hand-crafted features to perform clinically-relevant retrieval of diabetic retinopathy images. The second task uses latent features to perform content-adaptive image enhancement. The third task ranks a pair of images based on their aestheticism. The goal of the last task is to capture localized image artifacts in small datasets with patch-level labels. For both these tasks, I propose novel deep architectures and show significant improvement over the previous state-of-art approaches. A suitable combination of feature representations augmented with an appropriate learning approach can increase performance for most visual computing tasks.
ContributorsChandakkar, Parag Shridhar (Author) / Li, Baoxin (Thesis advisor) / Yang, Yezhou (Committee member) / Turaga, Pavan (Committee member) / Davulcu, Hasan (Committee member) / Arizona State University (Publisher)
Created2017
Description
Mixture of experts is a machine learning ensemble approach that consists of individual models that are trained to be ``experts'' on subsets of the data, and a gating network that provides weights to output a combination of the expert predictions. Mixture of experts models do not currently see wide use due to difficulty in training diverse experts and high computational requirements. This work presents modifications of the mixture of experts formulation that use domain knowledge to improve training, and incorporate parameter sharing among experts to reduce computational requirements.
First, this work presents an application of mixture of experts models for quality robust visual recognition. First it is shown that human subjects outperform deep neural networks on classification of distorted images, and then propose a model, MixQualNet, that is more robust to distortions. The proposed model consists of ``experts'' that are trained on a particular type of image distortion. The final output of the model is a weighted sum of the expert models, where the weights are determined by a separate gating network. The proposed model also incorporates weight sharing to reduce the number of parameters, as well as increase performance.
Second, an application of mixture of experts to predict visual saliency is presented. A computational saliency model attempts to predict where humans will look in an image. In the proposed model, each expert network is trained to predict saliency for a set of closely related images. The final saliency map is computed as a weighted mixture of the expert networks' outputs, with weights determined by a separate gating network. The proposed model achieves better performance than several other visual saliency models and a baseline non-mixture model.
Finally, this work introduces a saliency model that is a weighted mixture of models trained for different levels of saliency. Levels of saliency include high saliency, which corresponds to regions where almost all subjects look, and low saliency, which corresponds to regions where some, but not all subjects look. The weighted mixture shows improved performance compared with baseline models because of the diversity of the individual model predictions.
First, this work presents an application of mixture of experts models for quality robust visual recognition. First it is shown that human subjects outperform deep neural networks on classification of distorted images, and then propose a model, MixQualNet, that is more robust to distortions. The proposed model consists of ``experts'' that are trained on a particular type of image distortion. The final output of the model is a weighted sum of the expert models, where the weights are determined by a separate gating network. The proposed model also incorporates weight sharing to reduce the number of parameters, as well as increase performance.
Second, an application of mixture of experts to predict visual saliency is presented. A computational saliency model attempts to predict where humans will look in an image. In the proposed model, each expert network is trained to predict saliency for a set of closely related images. The final saliency map is computed as a weighted mixture of the expert networks' outputs, with weights determined by a separate gating network. The proposed model achieves better performance than several other visual saliency models and a baseline non-mixture model.
Finally, this work introduces a saliency model that is a weighted mixture of models trained for different levels of saliency. Levels of saliency include high saliency, which corresponds to regions where almost all subjects look, and low saliency, which corresponds to regions where some, but not all subjects look. The weighted mixture shows improved performance compared with baseline models because of the diversity of the individual model predictions.
ContributorsDodge, Samuel Fuller (Author) / Karam, Lina (Thesis advisor) / Jayasuriya, Suren (Committee member) / Li, Baoxin (Committee member) / Turaga, Pavan (Committee member) / Arizona State University (Publisher)
Created2018
Description
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.
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.
ContributorsChandran, Sreenithy (Author) / Jayasuriya, Suren (Thesis advisor) / Turaga, Pavan (Committee member) / Dasarathy, Gautam (Committee member) / Arizona State University (Publisher)
Created2019
Description
The human motion is defined as an amalgamation of several physical traits such as bipedal locomotion, posture and manual dexterity, and mental expectation. In addition to the “positive” body form defined by these traits, casting light on the body produces a “negative” of the body: its shadow. We often interchangeably use with silhouettes in the place of shadow to emphasize indifference to interior features. In a manner of speaking, the shadow is an alter ego that imitates the individual.
The principal value of shadow is its non-invasive behaviour of reflecting precisely the actions of the individual it is attached to. Nonetheless we can still think of the body’s shadow not as the body but its alter ego.
Based on this premise, my thesis creates an experiential system that extracts the data related to the contour of your human shape and gives it a texture and life of its own, so as to emulate your movements and postures, and to be your extension. In technical terms, my thesis extracts abstraction from a pre-indexed database that could be generated from an offline data set or in real time to complement these actions of a user in front of a low-cost optical motion capture device like the Microsoft Kinect. This notion could be the system’s interpretation of the action which creates modularized art through the abstraction’s ‘similarity’ to the live action.
Through my research, I have developed a stable system that tackles various connotations associated with shadows and the need to determine the ideal features that contribute to the relevance of the actions performed. The implication of Factor Oracle [3] pattern interpretation is tested with a feature bin of videos. The system also is flexible towards several methods of Nearest Neighbours searches and a machine learning module to derive the same output. The overall purpose is to establish this in real time and provide a constant feedback to the user. This can be expanded to handle larger dynamic data.
In addition to estimating human actions, my thesis best tries to test various Nearest Neighbour search methods in real time depending upon the data stream. This provides a basis to understand varying parameters that complement human activity recognition and feature matching in real time.
The principal value of shadow is its non-invasive behaviour of reflecting precisely the actions of the individual it is attached to. Nonetheless we can still think of the body’s shadow not as the body but its alter ego.
Based on this premise, my thesis creates an experiential system that extracts the data related to the contour of your human shape and gives it a texture and life of its own, so as to emulate your movements and postures, and to be your extension. In technical terms, my thesis extracts abstraction from a pre-indexed database that could be generated from an offline data set or in real time to complement these actions of a user in front of a low-cost optical motion capture device like the Microsoft Kinect. This notion could be the system’s interpretation of the action which creates modularized art through the abstraction’s ‘similarity’ to the live action.
Through my research, I have developed a stable system that tackles various connotations associated with shadows and the need to determine the ideal features that contribute to the relevance of the actions performed. The implication of Factor Oracle [3] pattern interpretation is tested with a feature bin of videos. The system also is flexible towards several methods of Nearest Neighbours searches and a machine learning module to derive the same output. The overall purpose is to establish this in real time and provide a constant feedback to the user. This can be expanded to handle larger dynamic data.
In addition to estimating human actions, my thesis best tries to test various Nearest Neighbour search methods in real time depending upon the data stream. This provides a basis to understand varying parameters that complement human activity recognition and feature matching in real time.
ContributorsSeshasayee, Sudarshan Prashanth (Author) / Sha, Xin Wei (Thesis advisor) / Turaga, Pavan (Thesis advisor) / Tinapple, David A (Committee member) / Arizona State University (Publisher)
Created2016
Description
High-level inference tasks in video applications such as recognition, video retrieval, and zero-shot classification have become an active research area in recent years. One fundamental requirement for such applications is to extract high-quality features that maintain high-level information in the videos.
Many video feature extraction algorithms have been purposed, such as STIP, HOG3D, and Dense Trajectories. These algorithms are often referred to as “handcrafted” features as they were deliberately designed based on some reasonable considerations. However, these algorithms may fail when dealing with high-level tasks or complex scene videos. Due to the success of using deep convolution neural networks (CNNs) to extract global representations for static images, researchers have been using similar techniques to tackle video contents. Typical techniques first extract spatial features by processing raw images using deep convolution architectures designed for static image classifications. Then simple average, concatenation or classifier-based fusion/pooling methods are applied to the extracted features. I argue that features extracted in such ways do not acquire enough representative information since videos, unlike images, should be characterized as a temporal sequence of semantically coherent visual contents and thus need to be represented in a manner considering both semantic and spatio-temporal information.
In this thesis, I propose a novel architecture to learn semantic spatio-temporal embedding for videos to support high-level video analysis. The proposed method encodes video spatial and temporal information separately by employing a deep architecture consisting of two channels of convolutional neural networks (capturing appearance and local motion) followed by their corresponding Fully Connected Gated Recurrent Unit (FC-GRU) encoders for capturing longer-term temporal structure of the CNN features. The resultant spatio-temporal representation (a vector) is used to learn a mapping via a Fully Connected Multilayer Perceptron (FC-MLP) to the word2vec semantic embedding space, leading to a semantic interpretation of the video vector that supports high-level analysis. I evaluate the usefulness and effectiveness of this new video representation by conducting experiments on action recognition, zero-shot video classification, and semantic video retrieval (word-to-video) retrieval, using the UCF101 action recognition dataset.
Many video feature extraction algorithms have been purposed, such as STIP, HOG3D, and Dense Trajectories. These algorithms are often referred to as “handcrafted” features as they were deliberately designed based on some reasonable considerations. However, these algorithms may fail when dealing with high-level tasks or complex scene videos. Due to the success of using deep convolution neural networks (CNNs) to extract global representations for static images, researchers have been using similar techniques to tackle video contents. Typical techniques first extract spatial features by processing raw images using deep convolution architectures designed for static image classifications. Then simple average, concatenation or classifier-based fusion/pooling methods are applied to the extracted features. I argue that features extracted in such ways do not acquire enough representative information since videos, unlike images, should be characterized as a temporal sequence of semantically coherent visual contents and thus need to be represented in a manner considering both semantic and spatio-temporal information.
In this thesis, I propose a novel architecture to learn semantic spatio-temporal embedding for videos to support high-level video analysis. The proposed method encodes video spatial and temporal information separately by employing a deep architecture consisting of two channels of convolutional neural networks (capturing appearance and local motion) followed by their corresponding Fully Connected Gated Recurrent Unit (FC-GRU) encoders for capturing longer-term temporal structure of the CNN features. The resultant spatio-temporal representation (a vector) is used to learn a mapping via a Fully Connected Multilayer Perceptron (FC-MLP) to the word2vec semantic embedding space, leading to a semantic interpretation of the video vector that supports high-level analysis. I evaluate the usefulness and effectiveness of this new video representation by conducting experiments on action recognition, zero-shot video classification, and semantic video retrieval (word-to-video) retrieval, using the UCF101 action recognition dataset.
ContributorsHu, Sheng-Hung (Author) / Li, Baoxin (Thesis advisor) / Turaga, Pavan (Committee member) / Liang, Jianming (Committee member) / Tong, Hanghang (Committee member) / Arizona State University (Publisher)
Created2016