Benjamin Z. Yao

I2T: Image Parsing to Text Description

In this paper, we present an image parsing to text description (I2T) framework that generates text descriptions of image and video content based on image understanding. The proposed I2T framework follows three steps: 1) input images (or video frames) are decomposed into their constituent visual patterns by an image parsing engine, in a spirit similar to parsing sentences in natural language; 2) the image parsing results are converted into semantic representation in the form of Web ontology language (OWL), which enables seamless integration with general knowledge bases; and 3) a text generation engine converts the results from previous steps into semantically meaningful, human readable, and query-able text reports. The centerpiece of the I2T framework is an and-or graph (AoG) visual knowledge representation, which provides a graphical representation serving as prior knowledge for representing diverse visual patterns and provides top-down hypotheses during the image parsing. The AoG embodies vocabularies of visual elements including primitives, parts, objects, scenes as well as a stochastic image grammar that specifies syntactic relations (i.e., compositional) and semantic relations (e.g., categorical, spatial, temporal, and functional) between these visual elements. Therefore, the AoG is a unified model of both categorical and symbolic representations of visual knowledge. The proposed I2T framework has two objectives. First, we use semiautomatic method to parse images from the Internet in order to build an AoG for visual knowledge representation. Our goal is to make the parsing process more and more automatic using the learned AoG model. Second, we use automatic methods to parse image/video in specific domains and generate text reports that are useful for real-world applications. In the case studies at the end of this paper, we demonstrate two automatic I2T systems: a maritime and urban scene video surveillance system and a real-time automatic driving scene understanding system.

Unsupervised learning of event AND-OR grammar and semantics from video

We study the problem of automatically learning event AND-OR grammar from videos of a certain environment, e.g. an office where students conduct daily activities. We propose to learn the event grammar under the information projection and minimum description length principles in a coherent probabilistic framework, without manual supervision about what events happen and when they happen. Firstly a predefined set of unary and binary relations are detected for each video frame: e.g. agents position, pose and interaction with environment. Then their co-occurrences are clustered into a dictionary of simple and transient atomic actions. Recursively these actions are grouped into longer and complexer events, resulting in a stochastic event grammar. By modeling time constraints of successive events, the learned grammar becomes context-sensitive. We introduce a new dataset of surveillance-style video in office, and present a prototype system for video analysis integrating bottom-up detection, grammatical learning and parsing. On this dataset, the learning algorithm is able to automatically discover important events and construct a stochastic grammar, which can be used to accurately parse newly observed video. The learned grammar can be used as a prior to improve the noisy bottom-up detection of atomic actions. It can also be used to infer semantics of the scene. In general, the event grammar is an efficient way for common knowledge acquisition from video.

Learning deformable action templates from cluttered videos

In this paper, we present a Deformable Action Template (DAT) model that is learnable from cluttered real-world videos with weak supervisions. In our generative model, an action template is a sequence of image templates each of which consists of a set of shape and motion primitives (Gabor wavelets and optical-flow patches) at selected orientations and locations. These primitives are allowed to slightly perturb their locations and orientations to account for spatial deformations. We use a shared pursuit algorithm to automatically discover a best set of primitives and weights by maximizing the likelihood over one or more aligned training examples. Since it is extremely hard to accurately label human actions from real-world videos, we use a three-step semi-supervised learning procedure. 1) For each human action class, a template is initialized from a labeled (one bounding-box per frame) training video. 2) The template is used to detect actions from other training videos of the same class by a dynamic space-time warping algorithm, which searches a best match between the template and target video in 5D space (x, y, scale, ttempiate and ttarget) using dynamic programming. 3) The template is updated by the shared pursuit algorithm over all aligned videos. The 2nd and 3rd steps iterate several times to arrive at an optimal action template. We tested our algorithm on a cluttered action dataset (the CMU dataset) and achieved favorable performance than [7]. Our classification performance on the KTH dataset is also comparable to state-of-the-arts.

A hierarchical and contextual model for aerial image understanding

In this paper we present a novel method for parsing aerial images with a hierarchical and contextual model learned in a statistical framework. We learn hierarchies at the scene and object levels to handle the difficult task of representing scene elements at different scales and add contextual constraints to resolve ambiguities in the scene interpretation. This allows the model to rule out inconsistent detections, like cars on trees, and to verify low probability detections based on their local context, such as small cars in parking lots. We also present a two-step algorithm for parsing aerial images that first detects object-level elements like trees and parking lots using color histograms and bag-of-words models, and objects like roofs and roads using compositional boosting, a powerful method for finding image structures. We then activate the top-down scene model to prune false positives from the first stage. We learn this scene model in a minimax entropy framework and show unique samples from our prior model, which capture the layout of scene objects. We present experiments showing that hierarchical and contextual information greatly reduces the number of false positives in our results.

Animated Pose Templates for Modeling and Detecting Human Actions

This paper presents animated pose templates (APTs) for detecting short-term, long-term, and contextual actions from cluttered scenes in videos. Each pose template consists of two components: 1) a shape template with deformable parts represented in an And-node whose appearances are represented by the Histogram of Oriented Gradient (HOG) features, and 2) a motion template specifying the motion of the parts by the Histogram of Optical-Flows (HOF) features. A shape template may have more than one motion template represented by an Or-node. Therefore, each action is defined as a mixture (Or-node) of pose templates in an And-Or tree structure. While this pose template is suitable for detecting short-term action snippets in two to five frames, we extend it in two ways: 1) For long-term actions, we animate the pose templates by adding temporal constraints in a Hidden Markov Model (HMM), and 2) for contextual actions, we treat contextual objects as additional parts of the pose templates and add constraints that encode spatial correlations between parts. To train the model, we manually annotate part locations on several keyframes of each video and cluster them into pose templates using EM. This leaves the unknown parameters for our learning algorithm in two groups: 1) latent variables for the unannotated frames including pose-IDs and part locations, 2) model parameters shared by all training samples such as weights for HOG and HOF features, canonical part locations of each pose, coefficients penalizing pose-transition and part-deformation. To learn these parameters, we introduce a semi-supervised structural SVM algorithm that iterates between two steps: 1) learning (updating) model parameters using labeled data by solving a structural SVM optimization, and 2) imputing missing variables (i.e., detecting actions on unlabeled frames) with parameters learned from the previous step and progressively accepting high-score frames as newly labeled examples. This algorithm belongs to a family of optimization methods known as the Concave-Convex Procedure (CCCP) that converge to a local optimal solution. The inference algorithm consists of two components: 1) Detecting top candidates for the pose templates, and 2) computing the sequence of pose templates. Both are done by dynamic programming or, more precisely, beam search. In experiments, we demonstrate that this method is capable of discovering salient poses of actions as well as interactions with contextual objects. We test our method on several public action data sets and a challenging outdoor contextual action data set collected by ourselves. The results show that our model achieves comparable or better performance compared to state-of-the-art methods.

Learning and parsing video events with goal and intent prediction

In this paper, we present a framework for parsing video events with stochastic Temporal And-Or Graph (T-AOG) and unsupervised learning of the T-AOG from video. This T-AOG represents a stochastic event grammar. The alphabet of the T-AOG consists of a set of grounded spatial relations including the poses of agents and their interactions with objects in the scene. The terminal nodes of the T-AOG are atomic actions which are specified by a number of grounded relations over image frames. An And-node represents a sequence of actions. An Or-node represents a number of alternative ways of such concatenations. The And-Or nodes in the T-AOG can generate a set of valid temporal configurations of atomic actions, which can be equivalently represented as the language of a stochastic context-free grammar (SCFG). For each And-node we model the temporal relations of its children nodes to distinguish events with similar structures but different temporal patterns and interpolate missing portions of events. This makes the T-AOG grammar context-sensitive. We propose an unsupervised learning algorithm to learn the atomic actions, the temporal relations and the And-Or nodes under the information projection principle in a coherent probabilistic framework. We also propose an event parsing algorithm based on the T-AOG which can understand events, infer the goal of agents, and predict their plausible intended actions. In comparison with existing methods, our paper makes the following contributions. (i) We represent events by a T-AOG with hierarchical compositions of events and the temporal relations between the sub-events. (ii) We learn the grammar, including atomic actions and temporal relations, automatically from the video data without manual supervision. (iii) Our algorithm infers the goal of agents and predicts their intents by a top-down process, handles events insertion and multi-agent events, keeps all possible interpretations of the video to preserve the ambiguities, and achieves the globally optimal parsing solution in a Bayesian framework. (iv) The algorithm uses event context to improve the detection of atomic actions, segment and recognize objects in the scene. Extensive experiments, including indoor and out door scenes, single and multiple agents events, are conducted to validate the effectiveness of the proposed approach.

Learning a scene contextual model for tracking and abnormality detection

In this paper we present a novel framework for learning contextual motion model involving multiple objects in far-field surveillance video and apply the learned model to improving the performance of objects tracking and abnormal event detection. We represent trajectory of multiple objects by a 3D graph G in x,y,t, which is augmented by a number of spatio-temporal relations (links) between moving and static objects in the scene (e.g. relation between crosswalk, pedestrian and car). An inhomogeneous Markov model p is defined over G, whose parameters are estimated by MLE method and relations are pursued by a minimax entropy principle (as in texture modeling) [16] so that we can synthesize entirely new video sequences that reproduce the observed statistics from training video. With the learned model, we define the abnormality of a subgraph given its neighborhood by log-likelihood ratio test, which is estimated by importance sampling. The learned model is applied to tracking and abnormal event detection. Our experiments show that the learned model improve tracking performance and detect sophisticated abnormal events like traffic rule violation.

Learning Compositional Models for Object Categories From Small Sample Sets

In this chapter we present a method for learning a compositional model in a minimax entropy framework for modeling object categories with large intra-class variance. The model we learn incorporates the flexibility of a stochastic context free grammar (SCFG) to account for the variation in object structure with the neighborhood constraints of a Markov random field (MRF) to enforce spatial context. We learn the model through a generalized minimax entropy framework that accounts for the dynamic structure of the hierarchical model. We first learn the SCFG parameters using the frequencies of object parts, then pursue spatial relations in order of greatest information gain. The learned model can generalize from a small set of training samples (n < 100) to generate a combinatorially large number of novel instances using stochastic sampling. To verify our learning method and model performance, we present plots of KL divergence minimization as the algorithm proceeds, and show that samples from the model become more realistic as more spatial relations are added. We also show the model accurately predicting missing or undetected parts for top-down recognition along with preliminary results showing that the model can learn a large space of category appearances from a very small (n < 15) number of training samples. This process is similar to “recognition-by-components”, a theory that postulates that biological vision systems recognize objects as composed from a dictionary of commonly appearing 3D structures. Finally, we discuss a compositional boosting algorithm for inference and show examples using it for object recognition. This article is a chapter from the book Object Categorization: Computer and Human Vision Perspectives, edited by Sven Dickinson, Ales Leonardis, Bernt Schiele, and Michael J. Tarr (Cambridge University Press). University of California Los Angeles, Los Angeles, CA. Lotus Hill Research Institute, EZhou, China.

Reconfigurable templates for robust vehicle detection and classification

In this paper, we learn a reconfigurable template for detecting vehicles and classifying their types. We adopt a popular design for the part based model that has one coarse template covering entire object window and several small high-resolution templates representing parts. The reconfigurable template can learn part configurations that capture the spatial correlation of features for a deformable part based model. The features of templates are Histograms of Gradients (HoG). In order to better describe the actual dimensions and locations of “parts” (i.e. features with strong spatial correlations), we design a dictionary of rectangular primitives of various sizes, aspect-ratios and positions. A configuration is defined as a subset of non-overlapping primitives from this dictionary. To learn the optimal configuration using SVM amounts, we need to find the subset of parts that minimize the regularized hinge loss, which leads to a non-convex optimization problem. We solve this problem by replacing the hinge loss with a negative sigmoid loss that can be approximately decomposed into losses (or negative sigmoid scores) of individual parts. In the experiment, we compare our method empirically with group lasso and a state of the art method [7] and demonstrate that models learned with our method outperform others on two computer vision applications: vehicle localization and vehicle model recognition.

Inferring social roles in long timespan video sequence

In this paper, we present a method for inferring social roles of agents (persons) from their daily activities in long surveillance video sequences. We define activities as interactions between an agents position and semantic hotspots within the scene. Given a surveillance video, our method first tracks the locations of agents then automatically discovers semantic hotspots in the scene. By enumerating spatial/temporal locations between an agents feet and hotspots in a scene, we define a set of atomic actions, which in turn compose sub-events and events. The numbers and types of events performed by an agent are assumed to be driven by his/her social role. With the grammar model induced by composition rules, an adapted Earley parser algorithm is used to parse the trajectories into events, sub-events and atomic actions. With probabilistic output of events, the roles of agents can be predicted under the Bayesian inference framework. Experiments are carried out on a challenging 8.5 hours video from a surveillance camera in the lobby of a research lab. The video contains 7 different social roles including “manager”, “researcher”, “developer”, “engineer”, “staff”, “visitor” and “mailman”. Results show that our proposed method can predict the role of each agent with high precision.