Wang Kongqiao

Feasibility of building robust surface electromyography-based hand gesture interfaces

This study explored the feasibility of building robust surface electromyography (EMG)-based gesture interfaces starting from the definition of input command gestures. As a first step, an offline experimental scheme was carried out for extracting user-independent input command sets with high class separability, reliability and low individual variations from 23 classes of hand gestures. Then three types (same-user, multi-user and cross-user test) of online experiments were conducted to demonstrate the feasibility of building robust surface EMG-based interfaces with the hand gesture sets recommended by the offline experiments. The research results reported in this paper are useful for the development and popularization of surface EMG-based gesture interaction technology.

Exploration on the feasibility of building muscle-computer interfaces using neck and shoulder motions

This paper investigates the feasibility of building muscle-computer interfaces starting from surface Electromyography (SEMG) -based neck and shoulder motion recognition. In order to reach the research goal, a real-time SEMG sensing, processing and classification system was developed firstly. Then two types of SEMG recognition experiments, namely user-specific and user-independent classification, were designed and conducted on seven kinds of neck and shoulder motions to explore the feasibility of using these motions as input commands of muscle-computer interfaces. In all 9 subjects took part in these experiments, 97.8% and 84.6% overall average recognition accuracies were obtained in user-specific and user-independent experiments respectively. The experimental results demonstrate that it is possible to build muscle-computer interfaces with neck and shoulder motions. In addition, the results of cross-time experiments designed to explore the relationship between training and accuracy in user-specific recognition indicate that users can interact accurately with computers using the defined motions only after four times training in different days.

Test-Retest Repeatability of Surface Electromyography Measurement for Hand Gesture

This study explores the test-retest repeatability of surface EMG measurements during hand gesture tasks across days and across subjects. Subjects took part in the data collection experiments on five separate days in a four-week time period. Surface EMG (sEMG) data was collected from the forearm. Intrasubject and intersubject test-retest repeatability of mean absolute values (MAV) and four-order AR model coefficients derived from 8-channel sEMG measurements were investigated using the coefficients of multiple correlation (CMCs) and coefficients of variation (CVs). Experimental results indicate moderate to high test-retest reliability for MAV and AR coefficients of sEMG measurements for nine of the ten studied hand gestures. The differences in test-retest repeatability among the hand gesture tasks also provide some insight into the effects of individual differences, electrode placement and gesture-specific characteristic of sEMG measurement.

Local binary pattern probability model based facial feature localization

In this paper, an active shape model (ASM) based facial feature localization strategy is proposed, which employs a local binary pattern (LBP) probability model. Due to the computation simplicity and illumination insensitivity of LBP texture descriptor and the learning ability of the probability model, the algorithm is robust and fast. In addition, component-based ASM is used to impose reasonable constraints on the shape. Multi-state shape and texture models with state classifier are trained to handle highly flexible components, i.e. eyes and mouth. Our database consisting of tens of persons with various expressions and illuminations is used to train and verify the proposed algorithm. The experiments demonstrate its accuracy, efficiency and robustness.

Dynamic gesture recognition based on multiple sensors fusion technology

This paper investigates the roles of a three-axis accelerometer, surface electromyography sensors and a webcam for dynamic gesture recognition. A decision-level multiple sensor fusion method based on action elements is proposed to distinguish a set of 20 kinds of dynamic hand gestures. Experiments are designed and conducted to collect three kinds of sensor data stream simultaneously during gesture implementation and compare the performance of different subsets in gesture recognition. Experimental results from three subjects show that the combination of three kinds of sensor achieves recognition accuracies at 87.5%-91.8%, which are higher largely than that of the single sensor conditions. This study is valuable to realize continuous and dynamic gesture recognition based on multiple sensor fusion technology for multi-model interaction.

Automatical face detection in images with complex background

In this paper, we presented a robust coarse-to-fine detection engine. In this engine, the structure based scheme is used to detect all face candidates, while the statistic based scheme to verify the detected candidates. All concave horizontal features (CHFs) are extracted from images under multi-scales and multi-levels to create a list of CHF images at the first stage. At the second stage, the face candidates are detected on the CHF images based on the trigonal relation between the eye pairs and mouths of human faces. The last stage is the face verification, which check out all real faces from the candidates. This robust face detection method is fast, and insensitive to facial poses, facial scales, uneven lighting, and etc. it can challenge any other face detection methods.

A Method of Representing Videos Based on User Interest Model in Mobile Multimedia Service

With the arrival of 3G era, video representation based on user interest has been an active research area in the past few years. Contrasting to the traditional approaches, which merely involved the interest of most users (e. g. most visited videos on Youtube), this paper proposes a novel video representing method based on individual interest model which is formed by collecting and analyzing user behaviors. Furthermore, this model can be extended into group interest model by introducing multi-dimension homo-interest sequence. Experiment results from 100 user cases and database of 25 videos show the effectiveness of our method.