Gesture sequence recognition with one shot learned CRF/HMM hybrid model

基于深度学习CRF/HMM混合模型的手势序列识别

Image and Vision Computing, Volume 61, May 2017, Pages 12-21

Selma Belgacem, Clément Chatelain, Thierry Paquet

Abstract: In this paper, we propose a novel markovian hybrid system CRF/HMM for gesture recognition, and a novel motion description method called gesture signature for gesture characterisation. The gesture signature is computed using the optical flows in order to describe the location, velocity and orientation of the gesture global motion. We elaborated the proposed hybrid CRF/HMM model by combining the modeling ability of Hidden Markov Models and the discriminative ability of Conditional Random Fields. In the context of one-shot-learning, this model is applied to the recognition of gestures in videos. In this extreme case, the proposed framework achieves very interesting performance and remains independent from the moving object type, which suggest possible application to other motion-based recognition tasks.

Centroid tracking based dynamic hand gesture recognition using discrete Hidden Markov Models

离散隐马尔可夫模型（HMM）在基于形心跟踪动态手势识别中的应用

Neurocomputing, Volume 228, 8 March 2017, Pages 79-83

Prashan Premaratne, Shuai Yang, Peter Vial, Zubair Ifthikar

Abstract: In many dynamic hand gesture recognition contexts, time information is not adequately used. The extracted features of dynamic gestures usually do not carry explicit information about time in gesture classification. This results in under-utilized data for more important accurate classification. Another disadvantage is that the gesture classification is then confined to only simple gestures. We have overcome these limitations by introducing centroid tracking of hand gestures that captures and retains the time sequence information for feature extraction. This simplifies the classification of dynamic gestures as movement in time helps efficient classification without burdensome processing.

A dynamic gesture recognition and prediction system using the convexity approach

基于凸率分析法的动态手势识别与预测系统

Computer Vision and Image Understanding, Volume 155, February 2017, Pages 139-149

Pablo Barros, Nestor T. Maciel-Junior, Bruno J.T. Fernandes, Byron L.D. Bezerra, Sergio M.M. Fernandes

Abstract: Several researchers around the world have studied gesture recognition, but most of the recent techniques fall in the curse of dimensionality and are not useful in real time environment. This study proposes a system for dynamic gesture recognition and prediction using an innovative feature extraction technique, called the Convexity Approach. The proposed method generates a smaller feature vector to describe the hand shape with a minimal amount of data. For dynamic gesture recognition and prediction, the system implements two independent modules based on Hidden Markov Models and Dynamic Time Warping. Two experiments, one for gesture recognition and another for prediction, are executed in two different datasets, the RPPDI Dynamic Gestures Dataset and the Cambridge Hand Data, and the results are showed and discussed.

Using data dimensionality reduction for recognition of incomplete dynamic gestures

数据降维技术在不完全动态手势识别中的应用

Pattern Recognition Letters, In Press, Corrected Proof, Available online 6 January 2017

Miguel Simão, Pedro Neto, Olivier Gibaru

Abstract:Continuous gesture spotting is a major topic in human-robot interaction (HRI) research. Human gestures are captured by sensors that provide large amounts of data that can be redundant or incomplete, correlated or uncorrelated. Data dimensionality reduction (DDR) techniques allow to represent such data in a low-dimensional space, making the classification process more efficient. This study demonstrates that DDR can improve the classification accuracy and allows the classification of gesture patterns with incomplete data, i.e., with the initial 25%, 50% or 75% of data representing a given dynamic gesture (DG) - time series of positional and hand shape data. Re-sampling raw data with bicubic interpolation and principal component analysis (PCA) were used as DDR methods. The performance of different classifiers is compared in the classification 95 different signs of the UCI Australian Sign Language (High Quality) Dataset. Experimental tests indicate that the use of PCA-based features result in a classification accuracy that is higher with 25% of gesture data (93% accuracy) than with 100% of gesture data (82% accuracy). These results were obtained from a non-trained data set and the recognized gestures are used to control a robot in an collaborative process.