Incremental learning of human behaviors using hierarchical hidden Markov models

Abstract

Abstract-This paper proposes a novel approach for extracting a model of movement primitives and their sequential relationships during online observation of human motion. In the proposed approach, movement primitives, modeled as hidden Markov models, are autonomously segmented and learned incrementally during observation. At the same time, a higher abstraction level hidden Markov model is also learned, encapsulating the relationship between the movement primitives. For the higher level model, each hidden state represents a motion primitive, and the observation function is based on the likelihood that the observed data is generated by the motion primitive model. An approach for incremental training of the higher order model during online observation is developed. The approach is validated on a dataset of continuous movement data.

The Language of Actions: Recovering the Syntax and Semantics of Goal-Directed Human Activities

“…In recent years, sequence modeling tools have received an increasing amount of attention, as shown, for instance, in the survey by Weinland et al [24], in part because of their ability to model the temporal structure of actions at multiple levels of granularity. HMMs are a particularly popular framework as they have been shown to work well not only for the recognition of events but also for the parsing and segmentation of videos [10] with applications ranging from sign language understanding [6,14] to the evaluation of motor skills including the training of surgeons [26]. In the context of the recognition of human actions in video, Chen and Aggarval [5] use the output of an SVM to classify complete activities with HMMs, reaching a recognition accuracy of 90.9% on the KTH dataset.…”

“…As a result, behavioral analysis with HMMs is often done with motion capture data or other sensors [10] or, in the case of video-based action recognition, with object, hand and head trajectories [6,26]. This has typically forced researchers working with HMMs to work in controlled environments with restrictive setups.…”

“…The number of stages was determined by cross validation using n = [3,5,10,15] fixed stages. We also tested an adaptive number of stages depending on the mean frame length of the related unit and linear scaling (factor 10).…”

Simple and complex behavior learning using behavior hidden Markov model and CobART

“…In addition, Kulic and Nakamura [24] worked on an incremental model that can learn recognition of human behaviors by using observed data. In their approach, motion primitives are segmented by a modified version of Kohlmorgen and Lehm algorithm.…”

Learning the Combinatorial Structure of Demonstrated Behaviors with Inverse Feedback Control

“…For example, Kulic and Nakamura have proposed in [2] a method that first performs an unsupervised segmentation of the motion signal into small successive blocks (the segmentation technique itself is based on HMMs), and then performs clustering over HMM representations of each segmented block. Each group of similar motions is interpreted as a motor primitive.…”