Conditional models for contextual human motion recognition

Abstract

We present algorithms for recognizing human motion in monocular video sequences, based on discriminative Conditional Random Field (CRF) and Maximum Entropy Markov Models (MEMM). Existing approaches to this problem typically use generative (joint) structures like the Hidden Markov Model (HMM). Therefore they have to make simplifying, often unrealistic assumptions on the conditional independence of observations given the motion class labels and cannot accommodate overlapping features or long term contextual dependencies in the observation sequence. In contrast, conditional models like the CRFs seamlessly represent contextual dependencies, support efficient, exact inference using dynamic programming, and their parameters can be trained using convex optimization. We introduce conditional graphical models as complementary tools for human motion recognition and present an extensive set of experiments that show how these typically outperform HMMs in classifying not only diverse human activities like walking, jumping, running, picking or dancing, but also for discriminating among subtle motion styles like normal walk and wander walk.

A Study on Smoothing for Particle-Filtered 3D Human Body Tracking

“…A different approach is taken by Mün-dermann et al (2007) and Cheng and Trivedi (2007), who align a 3D visual hull body model to an observed visual hull constructed from silhouettes seen in multiple viewpoints in order to achieve viewpoint independence. Other researchers (Taycher et al 2006;Sminichisescu et al 2006) utilise statistical time-series models such as conditional random fields (CRFs) and maximum entropy Markov models (MEMMs) to perform tracking. However, failures in detecting the true limb or the full silhouette and the need to train limb detectors specific to the person being tracked means that discriminative approaches have difficulty with observation 'errors' (e.g.…”

A dynamic description logic based system for video event detection

“…Many efforts in video event detection have been done to overcome these challenges [1][2][3][4][5][6][7][8][9][10]. According to the proposed methods, a typical video event detection system can be divided into two components which are information abstraction component and event modeling component.…”

Locally time-invariant models of human activities using trajectories on the grassmannian

“…In contrast to the generative approaches discussed above, there exist discriminative approaches for modeling human actions. An in-depth discussion of discriminative models is beyond the scope of this paper, and we refer the reader to [17,10] and references therein.…”