We propose a modified dynamic time warping (DTW) algorithm that compares gesture-position sequences based on the direction of the gestural movement. Standard DTW does not specifically consider the two-dimensional characteristic of the user’s movement. Therefore, in gesture recognition, the sequence comparison by standard DTW needs to be improved. The proposed gesture-recognition system compares the sequences of the input gesture’s position with gesture positions saved in the database and selects the most similar gesture by filtering out unrelated gestures. The suggested algorithm uses the cosine similarity of the movement direction at each moment to calculate the difference and reflects the characteristics of the gesture movement by using the ratio of the Euclidean distance and the proportional distance to the calculated difference. Selective spline interpolation assists in solving the issue of recognition-decline at instances of gestures. Through experiments with public databases (MSRC-12 and G3D), the suggested algorithm revealed an improved performance on both databases compared to other methods.
In this research, we present a differential evolution approach to optimize the weights of dynamic time warping for multi-sensory based gesture recognition. Mainly, we aimed to develop a robust gesture recognition method that can be used in various environments. Both a wearable inertial sensor and a depth camera (Kinect Sensor) were used as heterogeneous sensors to verify and collect the data. The proposed approach was used for the calculation of optimal weight values and different characteristic features of heterogeneous sensor data, while having different effects during gesture recognition. In this research, we studied 27 different actions to analyze the data. As finding the optimal value of the data from numerous sensors became more complex, a differential evolution approach was used during the fusion and optimization of the data. To verify the performance accuracy of the presented method in this study, a University of Texas at Dallas Multimodal Human Action Datasets (UTD-MHAD) from previous research was used. However, the average recognition rates presented by previous research using respective methods were still low, due to the complexity in the calculation of the optimal values of the acquired data from sensors, as well as the installation environment. Our contribution was based on a method that enabled us to adjust the number of depth cameras and combine this data with inertial sensors (multi-sensors in this study). We applied a differential evolution approach to calculate the optimal values of the added weights. The proposed method achieved an accuracy 10% higher than the previous research results using the same database, indicating a much improved accuracy rate of motion recognition.
Cyber-physical systems, which closely integrate physical systems and humans, can be applied to a wider range of applications through user movement analysis. In three-dimensional (3D) gesture recognition, multiple sensors are required to recognize various natural gestures. Several studies have been undertaken in the field of gesture recognition; however, gesture recognition was conducted based on data captured from various independent sensors, which rendered the capture and combination of real-time data complicated. In this study, a 3D gesture recognition method using combined information obtained from multiple sensors is proposed. The proposed method can robustly perform gesture recognition regardless of a user’s location and movement directions by providing viewpoint-weighted values and/or motion-weighted values. In the proposed method, the viewpoint-weighted dynamic time warping with multiple sensors has enhanced performance by preventing joint measurement errors and noise due to sensor measurement tolerance, which has resulted in the enhancement of recognition performance by comparing multiple joint sequences effectively.
1 January 2016 in Contemporary Engineering Sciences
Dynamic time warping algorithm is a pattern matching algorithm that allows a nonlinear stretching of the data. In a recognition system using a matching algorithm, data clustering methods are used to reduce the number of gesture templates in the database, and thus reduce the computational cost; however, the recognition rate is degraded. In this paper, we proposed a DTW gesture recognition system that uses orientation histogram for feature extraction and candidate selection based on weighted probability histogram comparison. Probability histogram is obtained from orientation histogram. Experiment has shown that our method reduces the process time while keeping robustness against accuracy loss due to clustering method.
21 April 2016 in Advanced Science and Technology Letters