• Kim, Lee, and Essa (2011), “Gaussian Process Regression Flow for Analysis of Motion Trajectories,” in Proceedings of IEEE International Conference on Computer Vision (ICCV), 2011. [PDF] [WEBSITE] [VIDEO] [BIBTEX]

   @inproceedings{Kim2011-GPRF, Author = {K. Kim and D. Lee and I. Essa}, Booktitle = {Proceedings of IEEE International Conference on Computer Vision (ICCV)}, Month = {November}, Pdf = {http://www.cc.gatech.edu/~irfan/p/2011-Kim-GPRFAMT.pdf}, Publisher = {IEEE Computer Society}, Title = {Gaussian Process Regression Flow for Analysis of Motion Trajectories}, Url = {http://www.cc.gatech.edu/cpl/projects/gprf/}, Video = {http://www.youtube.com/watch?v=UtLr37hDQz0}, Year = {2011}}

Abstract

Analysis and Recognition of motions and activities of objects in videos requires effective representations for analysis and matching of motion trajectories. In this paper, we introduce a new representation specifically aimed at matching motion trajectories. We model a trajectory as a continuous dense flow field from a sparse set of vector sequences using Gaussian Process Regression. Furthermore, we introduce a random sampling strategy for learning stable classes of motions from limited data.

Our representation allows for incrementally predicting possible paths and detecting anomalous events from online trajectories. This representation also supports matching of complex motions with acceleration changes and pauses or stops within a trajectory. We use the proposed approach for classifying and predicting motion trajectories in traffic monitoring domains and test on several data sets. We show that our approach works well on various types of complete and incomplete trajectories from a variety of video data sets with different frame rates

http://www.youtube.com/watch?v=UtLr37hDQz0

Abstract

My research group is focused on a variety of approaches for (a) low-level video analysis and synthesis and (b) recognizing activities in videos. In this talk, I will concentrate on two of our recent efforts. One effort aimed at robust spatio-temporal segmentation of video and another on using motion and flow to recognize and predict actions from video.

In the first part of the talk, I will present an efficient and scalable technique for spatio-temporal segmentation of long video sequences using a hierarchical graph-based algorithm. In this work, we begin by over segmenting a volumetric video graph into space-time regions grouped by appearance. We then construct a “region graph” over the obtained segmentation and iteratively repeat this process over multiple levels to create a tree of spatio-temporal segmentations. This hierarchical approach generates high quality segmentations, which are temporally coherent with stable region boundaries, and allows subsequent applications to choose from varying levels of granularity. We further improve segmentation quality by using dense optical flow to guide temporal connections in the initial graph. I will demonstrate a variety of examples of how this robust segmentation works, and will show additional examples of video-retargeting that use spatio-temporal saliency derived from this segmentation approach. (Matthias Grundmann, Vivek Kwatra, Mei Han, Irfan Essa, CVPR 2010, in collaboration with Google Research).

In the second part of this talk, I will show that constrained multi-agent events can be analyzed and even predicted from video. Such analysis requires estimating the global movements of all players in the scene at any time, and is needed for modeling and predicting how the multi-agent play evolves over time on the playing field. To this end, we propose a novel approach to detect the locations of where the play evolution will proceed, e.g. where interesting events will occur, by tracking player positions and movements over time. To achieve this, we extract the ground level sparse movement of players in each time-step, and then generate a dense motion field. Using this field we detect locations where the motion converges, implying positions towards which the play is evolving. I will show examples of how we have tested this approach for soccer, basketball and hockey. (Kihwan Kim, Matthias Grundmann, Ariel Shamir, Iain Matthews, Jessica Hodgins, Irfan Essa, CVPR 2010, in collaboration with Disney Research).

Time permitting, I will show some more videos of our recent work on video analysis and synthesis. For more information, papers, and videos, see my website.

Friday September 24, 2010 from 11:00am to 12:00pm

Kitware has received a $13,883,314 contract from Defense Advanced Research Projects Agency (DARPA) to develop a software system capable of automatically and interactively discovering actionable intelligence from wide area motion imagery (WAMI) of complex urban, suburban, and rural environments.

The primary information elements in WAMI data are moving entities in the context of roads, buildings, and other scene features. These entities, while exploitable, often yield fragmented tracks in complex urban environments due to occlusions, stops, and other factors. Kitware’s software system will use algorithmic solutions to associate tracks and then identify and integrate local events to detect potential threats and perform forensic analysis.

The developed algorithms will form the basis of a software prototype called the Persistent Stare Exploitation and Analysis System (PerSEAS) that will significantly augment an end-user’s ability to discover novel intelligence using models of activities, normalcy, and context. Since the vast majority of events are normal and pose no threat, the models must cross-integrate singular events to discover relationships and anomalies that are indicative of suspicious behavior or match previously learned – or defined – threat activity.

The advanced PerSEAS system will markedly improve an analyst’s ability to handle burgeoning WAMI data and reduce the time required to perform many current exploitation tasks, greatly enhancing the military’s capability to analyze and utilize the data for forensic analysis and through the issuance of timely threat alerts with a minimal number of false alarms.

Due to the complex, multi-disciplinary nature of the research, Kitware will partner with academic experts in the fields of computer vision, probabilistic reasoning, machine learning and other related domains. Phase I of the research is expected to be completed in two years.

The awarded contract will expand Kitware’s leadership in the field of computer vision, video analysis and advanced visualization software. The project will build upon our previous DARPA-sponsored research into content-based video retrieval on the VIRAT program; anomaly detection on the PANDA program; and the recognition of complex multi-agent activities in video.

To meet the PerSEAS program’s needs, Kitware has assembled a world-class team including four leading defense technology companies, Northrop Grumman Corporation, ; Honeywell Automation and Control Solutions Laboratories, Aptima, Inc., and Navia, Inc. As well as multiple internationally-renowned research institutions, including: the University of California, Berkeley; Computer Vision Laboratory, University of Maryland; Rensselaer Polytechnic Institute; the Computer Vision Lab at the University of Central Florida; the School of Interactive Computing at Georgia Tech and its affiliated Center for Robotics & Intelligent Machines; and Columbia University.

Abstract

Videos of multi-player team sports provide a challenging domain for dynamic scene analysis. Player actions and interactions are complex as they are driven by many factors, such as the short-term goals of the individual player, the overall team strategy, the rules of the sport, and the current context of the game. We show that constrained multi-agent events can be analyzed and even predicted from video. Such analysis requires estimating the global movements of all players in the scene at any time, and is needed for modeling and predicting how the multi-agent play evolves over time on the field. To this end, we propose a novel approach to detect the locations of where the play evolution will proceed, e.g. where interesting events will occur, by tracking player positions and movements over time. We start by extracting the ground level sparse movement of players in each time-step, and then generate a dense motion field. Using this field we detect locations where the motion converges, implying positions towards which the play is evolving. We evaluate our approach by analyzing videos of a variety of complex soccer plays.

Abstract

We present a novel approach for robust localization of multiple people observed using multiple cameras. We usethis location information to generate sports visualizations,which include displaying a virtual offside line in soccer games, and showing players’ positions and motion patterns.Our main contribution is the modeling and analysis for the problem of fusing corresponding players’ positional informationas finding minimum weight K-length cycles in complete K-partite graphs. To this end, we use a dynamic programmingbased approach that varies over a continuum of being maximally to minimally greedy in terms of the numberof paths explored at each iteration. We present an end-to-end sports visualization framework that employs our proposed algorithm-class. We demonstrate the robustness of our framework by testing it on 60; 000 frames of soccerfootage captured over 5 different illumination conditions, play types, and team attire.

Part of the team led by Kitware Inc to work on Defense Advanced Research Projects Agency – Persistent Stare Exploitation and Analysis System (PerSEAS).

The Persistent Stare Exploitation and Analysis System (PerSEAS) program is developing the capability to automatically and interactively identify potential threats as they emerge based on the correlation of multiple disparate activities and events in wide area motion imagery (WAMI) and multi-INT data.  PerSEAS will enable new methods of threat hypothesis adjudication and forensic analysis through activity-based modeling and inferencing capabilities.

See

• DARPA BAA
• Wired Article on this project

Abstract

This paper investigates the recognition of human actions from three-dimensional (3-D) point clouds that encode the motions of people acting in sensor-distributed indoor environments. Data streams are time sequences of silhouettes extracted from cameras in the environment. From the 2-D silhouette contours we generate space–time streams by continuously aligning and stacking the contours along the time axis as third spatial dimension. The space–time stream of an observation sequence is segmented into parts corresponding to subactions using a pattern matching technique based on suffix trees and interval scheduling. Then, the segmented space–time shapes are processed by treating the shapes as 3-D point clouds and estimating global point feature histograms for them. The resultant models are clustered using statistical analysis and our experimental results indicate that the presented methods robustly derive different action classes. This holds despite large intra-class variance in the recorded datasets due to performances from different persons at different time intervals.

© Koninklijke Brill NV, Leiden and The Robotics Society of Japan, 2009

Overview of the approach.

Keywords: Action recognition, point cloud, global features, action segmentation

• Raffay Hamid, Siddhartha Maddi, Amos Johnson, Aaron Bobick, Irfan Essaand Charles Isbell (2009) “A novel sequence representation for unsupervised analysis of human activities” in Artificial Intelligence, Volume 173, Issue 14, September 2009, Pages 1221-1244. [PDF][DOI][Science Direct]

• Hamid, Maddi, Johnson, Bobick, Essa, and Isbell (2009), “A Novel Sequence Representation for Unsupervised Analysis of Human Activities,” Artificial Intelligence Journal, 2009. [BIBTEX]

  @article{2009-Hamid-NSRUAHA,
    Author = {R. Hamid and S. Maddi and A. Johnson and A. Bobick and I. Essa and C. Isbell},
    Date-Modified = {2011-12-08 21:27:48 +0000},
    Journal = {Artificial Intelligence Journal},
    Month = {May},
    Title = {A Novel Sequence Representation for Unsupervised Analysis of Human Activities},
    Year = {2009}}

Abstract

Formalizing computational models for everyday human activities remains an open challenge. Many previous approaches towards this end assume prior knowledge about the structure of activities, using which explicitly defined models are learned in a completely supervised manner. For a majority of everyday environments however, the structure of the in situ activities is generally not known a priori. In this paper we investigate knowledge representations and manipulation techniques that facilitate learning of human activities in a minimally supervised manner. The key contribution of this work is the idea that global structural information of human activities can be encoded using a subset of their local event subsequences, and that this encoding is sufficient for activity-class discovery and classification.

In particular, we investigate modeling activity sequences in terms of their constituent subsequences that we call event n-grams. Exploiting this representation, we propose a computational framework to automatically discover the various activity-classes taking place in an environment. We model these activity-classes as maximally similar activity-cliques in a completely connected graph of activities, and describe how to discover them efficiently. Moreover, we propose methods for finding characterizations of these discovered classes from a holistic as well as a by-parts perspective. Using such characterizations, we present a method to classify a new activity to one of the discovered activity-classes, and to automatically detect whether it is anomalous with respect to the general characteristics of its membership class. Our results show the efficacy of our approach in a variety of everyday environments.

Keywords: Temporal reasoning; Scene analysis; Computer vision

ABSTRACT

We propose the use of 3D (2D+time) Shape Context to recognize the spatial and temporal details inherent in human actions. We represent an action in a video sequence by a 3D point cloud extracted by sampling 2D silhouettes over time. A non-uniform sampling method is introduced that gives preference to fast moving body parts using a Euclidean 3D Distance Transform. Actions are then classified by matching the extracted point clouds. Our proposed approach is based on a global matching and does not require specific training to learn the model. We test the approach thoroughly on two publicly available datasets and compare to several state-of-the-art methods. The achieved classification accuracy is on par with or superior to the best results reported to date.

M. Raffay Hamid PhD (2008), “A Computational Framework For Unsupervised Analysis of Everyday Human Activities“, PhD Thesis, Georgia Institute of Techniology, College of Computing, Atlanta, GA. (Advisor: Aaron Bobick & Irfan Essa)

Abstract

In order to make computers proactive and assistive, we must enable them to perceive, learn, and predict what is happening in their surroundings. This presents us with the challenge of formalizing computational models of everyday human activities. For a majority of environments, the structure of the in situ activities is generally not known a priori. This thesis therefore investigates knowledge representations and manipulation techniques that can facilitate learning of such everyday human activities in a minimally supervised manner. 

A key step towards this end is finding appropriate representations for human activities. We posit that if we chose to describe activities as finite sequences of an appropriate set of events, then the global structure of these activities can be uniquely encoded using their local event sub-sequences. With this perspective at hand, we particularly investigate representations that characterize activities in terms of their fixed and variable length event subsequences. We comparatively analyze these representations in terms of their representational scope, feature cardinality and noise sensitivity.

Exploiting such representations, we propose a computational framework to discover the various activity-classes taking place in an environment. We model these activity-classes as maximally similar activity-cliques in a completely connected graph of activities, and describe how to discover them efficiently. Moreover, we propose methods for finding concise characterizations of these discovered activity-classes, both from a holistic as well as a by-parts perspective. Using such characterizations, we present an incremental method to classify

a new activity instance to one of the discovered activity-classes, and to automatically detect if it is anomalous with respect to the general characteristics of its membership class. Our results show the efficacy of our framework in a variety of everyday environments

• David Minnen PhD (2008): “Unsupervised Discovery of Activity Primitives from Multivariate Sensor Data“ Georgia Institute of Techniology, College of Computing, Atlanta, GA. (Advisors: Thad Starner & Irfan Essa)

Abstract

This research addresses the problem of temporal pattern discovery in real-valued, multivariate sensor data. Several algorithms were developed, and subsequent evaluation demonstrates that they can efficiently and accurately discover unknown recurring patterns in time series data taken from many different domains. Different data representations and motif models were investigated in order to design an algorithm with an improved balance between run-time and detection accuracy. The different data representations are used to quickly filter large data sets in order to detect potential patterns that form the basis of a more detailed analysis. The representations include global discretization, which can be efficiently analyzed using a suffix tree, local discretization with a corresponding random projection algorithm for locating similar pairs of subsequences, and a density-based detection method that operates on the original, real-valued data. In addition, a new variation of the multivariate motif discovery problem is proposed in which each pattern may span only a subset of the input features. An algorithm that can efficiently discover such “subdimensional” patterns was developed and evaluated. The discovery algorithms are evaluated by measuring the detection accuracy of discovered patterns relative to a set of expected patterns for each data set. The data sets used for evaluation are drawn from a variety of domains including speech, on-body inertial sensors, music, American Sign Language video, and GPS tracks.

• Minnen, Essa, Isbell, and Starner (2007), “Detecting Subdimensional Motifs: An Efficient Algorithm for Generalized Multivariate Pattern Discovery,” in Proceedings of IEEE International Conference on Data Mining (ICDM), 2007. [PDF] [DOI]
  [BIBTEX]

  @inproceedings{2007-Minnen-DSMEAGMPD,
    Author = {D. Minnen and I. Essa and C. Isbell and T. Starner},
    Booktitle = {Proceedings of IEEE International Conference on Data Mining (ICDM)},
    Doi = {10.1109/ICDM.2007.52},
    Month = {October},
    Pdf = {http://www.cc.gatech.edu/~irfan/p/2007-Minnen-DSMEAGMPD.pdf},
    Title = {Detecting Subdimensional Motifs: An Efficient Algorithm for Generalized Multivariate Pattern Discovery},
    Year = {2007}}

Abstract

Discovering recurring patterns in time series data is a fundamental problem for temporal data mining. This paper addresses the problem of locating subdimensional motifs in real-valued, multivariate time series, which requires the simultaneous discovery of sets of recurring patterns along with the corresponding relevant dimensions. While many approaches to motif discovery have been developed, most are restricted to categorical data, univariate time series, or multivariate data in which the temporal patterns span all of the dimensions. In this paper, we present an expected linear-time algorithm that addresses a generalization of multivariate pattern discovery in which each motif may span only a subset of the dimensions. To validate our algorithm, we discuss its theoretical properties and empirically evaluate it using several data sets including synthetic data and motion capture data collected by an on-body inertial sensor.

Abstract

Models of activity structure for unconstrained environments are generally not available a priori. Recent representational approaches to this end are limited by their computational complexity, and ability to capture activity structure only up to some fixed temporal scale. In this work, we propose Suffix Trees as an activity representation to efficiently extract structure of activities by analyzing their constituent event-subsequences over multiple temporal scales. We empirically compare Suffix Trees with some of the previous approaches in terms of feature cardinality, discriminative prowess, noise sensitivity and activity-class discovery. Finally, exploiting properties of Suffix Trees, we present a novel perspective on anomalous subsequences of activities, and propose an algorithm to detect them in linear-time. We present comparative results over experimental data, collected from a kitchen environment to demonstrate the competence of our proposed framework.

Abstract

We present an unsupervised framework to discover characterizations of everyday human activities, and demonstrate how such representations can be used to extract points of interest in event-streams. We begin with the usage of Suffix Trees as an efficient activity-representation to analyze the global structural information of activities, using their local event statistics over the entire continuum of their temporal resolution. Exploiting this representation, we discover characterizing event-subsequences and present their usage in an ensemble-based framework for activity classification. Finally, we propose a method to automatically detect subsequences of events that are locally atypical in a structural sense. Results over extensive data-sets, collected from multiple sensor-rich environments are presented, to show the competence and scalability of the proposed framework.

Minnen, D. Starner, T. Essa, I. Isbell, C.
College of Computing, Georgia Institute of Technology, Atlanta, GA 30332 USA.
This paper appears in: Wearable Computers, 2006 10th IEEE International Symposium on
Publication Date: Oct. 2006
On page(s): 11 – 18
Number of Pages: 11 – 18
Location: Montreux, Switzerland
ISSN: 1550-4816
ISBN: 1-4244-0598-x
Digital Object Identifier: 10.1109/ISWC.2006.286337
Posted online: 2007-01-22 09:58:15.0

Abstract

We present an approach to activity discovery, the unsupervised identification and modeling of human actions embedded in a larger sensor stream. Activity discovery can be seen as the inverse of the activity recognition problem. Rather than learn models from hand-labeled sequences, we attempt to discover motifs, sets of similar subsequences within the raw sensor stream, without the benefit of labels or manual segmentation. These motifs are statistically unlikely and thus typically correspond to important or characteristic actions within the activity. The problem of activity discovery differs from typicalmotif discovery, such as locating protein binding sites, because of the nature of time series data representing human activity. For example, in activity data, motifs will tend to be sparsely distributed, vary in length, and may only exhibit intra-motif similarity after appropriate time warping. In this paper, we motivate the activity discovery problem and present our approach for efficient discovery of meaningful actions from sensor data representing human activity. We empirically evaluate the approach on an exercise data set captured by a wrist-mounted, three-axis inertial sensor. Our algorithm successfully discovers motifs that correspond to the real exercises with a recall rate of 96.3% and overall accuracy of 86.7% over six exercises and 864 occurrences.

Abstract

Graphical models are often used to represent and recognize activities. Purely unsupervised methods (such as HMMs) can be trained automatically but yield models whose internal structure – the nodes – are difficult to interpret semantically. Manually constructed networks typically have nodes corresponding to sub-events, but the programming and training of these networks is tedious and requires extensive domain expertise. In this paper, we propose a semi-supervised approach in which a manually structured, Propagation Network (a form of a DBN) is initialized from a small amount of fully annotated data, and then refined by an EM-based learning method in an unsupervised fashion. During node refinement (the M step) a boosting-based algorithm is employed to train the evidence detectors of individual nodes. Experiments on a variety of data types – vision and inertial measurements – in several tasks demonstrate the ability to learn from as little as one fully annotated example accompanied by a small number of positive but non-annotated training examples. The system is applied to both recognition and anomaly detection tasks.

ABSTRACT

We present an approach for tracking varying number of objects through both temporally and spatially significant occlusions. Our method builds on the idea of object permanence to reason about occlusions. To this end, tracking is performed at both the region level and the object level. At the region level, a customized genetic algorithm is used to search for optimal region tracks. This limits the scope of object trajectories. At the object level, each object is located based on adaptive appearance models, spatial distributions and inter-occlusion relationships. The proposed architecture is capable of tracking objects even in the presence of long periods of full occlusions. We demonstrate the viability of this approach by experimenting on several videos of a user interacting with a variety of objects on a desktop.

Temporal Reasoning from Video to Temporal Synthesis of Video

Abstract

In this talk, I will present some ongoing work on extracting spatio-temporal cues from video for both synthesis of novel video sequences, and recognition of complex activities. I will start off with some of our earlier work on Video Textures, where repeating information is extracted to generate extended sequences of videos. I will then describe some of our extensions to this approach that allow for controlled generation of animations of video sprites. We have developed various learning and optimization techniques that allow for video-based animations of photo-realistic characters. Then I will describe our new approach for image and video synthesis that builds on optimal patch-based copying of samples. I will show how our method allows for iterative refinement and extends to synthesis of both images and video from very limited samples. In the next part of my talk, I will describe how a similar analysis of video can be used to recognize what a person is doing in a scene. Such an analysis of video, aimed at recognition, requires more contextual information about the environment. I will show how we leverage contextual information shared between actions and objects to recognize what is happening in complex environments. I will also show that by adding some form of grammar (we use Stochastic Context Free Grammar) we can recognize very complex, multi-tasked activities.

If time permits, I will describe (very briefly) the Aware Home project at Georgia Tech, which is one primary area of ongoing and future research for me and my group. Further information on my work with videos is available from my webpage at http://www.cc.gatech.edu/~irfan

Abstract

We present propagation networks (P-nets), a novel approach for representing and recognizing sequential activities that include parallel streams of action. We represent each activity using partially ordered intervals. Each interval is restricted by both temporal and logical constraints, including information about its duration and its temporal relationship with other intervals. P-nets associate one node with each temporal interval. Each node is triggered according to a probability density function that depends on the state of its parent nodes. Each node also has an associated observation function that characterizes supporting perceptual evidence. To facilitate real-time analysis, we introduce a particle filter framework to explore the conditional state space. We modify the original condensation algorithm to more efficiently sample a discrete state space (D-condensation). Experiments in the domain of blood glucose monitor calibration demonstrate both the representational power of P-nets and the effectiveness of the D-condensation algorithm.