Shuichi Enokida

Magnitude and phase spectra of foot motion for gait recognition

Magnitude and phase spectra of horizontal and vertical movement of ankles in a normal walk are effective and efficient signatures in gait recognition. An approach to use these spectra as phase-weighted magnitude spectra is also widely known. In this paper, we propose an integration of magnitude and phase spectra for gait recognition using AdaBoost classifier. At each round, a weak classifier evaluates each magnitude and phase spectra of a motion signal as dependent sub-features, then classification results of each sub-feature are normalized and summed for the final hypothesis output. Experimental results in same-day and cross-month tests with nine subjects show that using both magnitude and phase spectra improves the recognition results.

Gait recognition using spectral features of foot motion

Gait as a motion-based biometric has the merit of being non-contact and unobtrusive. In this paper, we proposed a gait recognition approach using spectral features of horizontal and vertical movement of ankles in a normal walk. Gait recognition experiments using the spectral features in term of the magnitude, phase and phase-weighted magnitude show that both magnitude and phase spectra are effective gait signatures, but magnitude spectra are slightly superior. We also proposed the use of geometrical mean based spectral features for gait recognition. Experimental results with 9 subjects show encouraging results in the same-day test, while the effect of time covariate is confirmed in the cross-month test.

Stochastic field model for autonomous robot learning

Through reinforcement learning, an autonomous robot creates an optimal policy which maps state space to action space. The mapping is obtained by trial and error through the interaction with a given environment. The mapping is represented as an action-value function. The environment accords an information in the form of scalar feedback known as a reinforcement signal. As a result of reinforcement learning, an action has the high action-value in each state. The optimal policy is equivalent to choosing an action which has the highest action-value in each state. Typically, even if an autonomous robot has continuous sensor values, the summation of discrete values is used as an action-value function to reduce learning time. However, the reinforcement learning algorithms including Q-learning suffer from errors due to state space sampling. To overcome the above, we propose an EQ-learning (extended Q-learning) based on a SFM (stochastic field model). EQ-learning is designed in order to accommodate continuous state space directly and to improve its generalization capability. Through EQ-learning, an action-value function is represented by the summation of weighted base functions, and an autonomous robot adjusts weights of base functions at learning stage. Other parameters (center coordinates, variance and so on) are adjusted at the unification stage where two similar functions are unified to a simpler function.

A Predictive Model for Gait Recognition

Gait Recognition has been paid an attention to as non-contact and unobtrusive biometric method. Magnitude and phase spectra of horizontal and vertical movement of ankles in a normal walk are effective and efficient signatures in gait recognition. However, gait recognition rate degrades significantly due to variance caused by covariates of clothing, surface or time lapse. In this paper, to improve gait recognition rate on a variety of footwear, a predictive model is proposed. The predictive model is able to estimate slipper gait from shoes gait. By using predictive slipper gait, much higher recognition rate is achieved for slipper gait over time lapse than ones without predictive model. The predictive model designed in this paper succeeds in separation of the variance due to a footwear covariate from the variance due to a time covariate.

Face Pose Estimation From Video Sequence Using Dynamic Bayesian Network

This paper describes a technique to estimate human face pose from color video sequence using dynamic Bayesian network(DBN). As face and facial features trackers usually track eyes, pupils, mouth corners and skin region(face), our proposed method utilizes merely three of these features - pupils, mouth center and skin region - to compute the evidence for DBN inference. No additional image processing algorithm is required, thus, it is simple and operates in real-time. The evidence, which are called horizontal ratio and vertical ratio in this paper, are determined using model-based technique and designed significantly to simultaneously solve two problems in tracking task; scaling factor and noise influence. Results reveal that the proposed method can be realized in real-time on a 2.2 GHz Celeron CPU machine with very satisfactory pose estimation results.

Extended Q-Learning: Reinforcement Learning Using Self-Organized State Space

We propose Extended Q-learning. To accommodate continuous state space directly and to improve its generalization capability. Through EQ-learning, an action-value function is represented by the summation of weighted base functions, and an autonomous robot adjusts weights of base functions at learning stage. Other parameters (center coordinates, variance and so on) are adjusted at unification stage where two similar functions are unified to a simpler function.

Safety Driving Assessment Based on Video Image Sequence Analysis

In this paper, a driving assessment mechanism based on video image sequence analysis techniques, such as traffic signs and road markings detection, and inter-vehicular distance estimation, is proposed. A smart device connected to 3G wireless networks is used to collect and send sensory information. These data are analyzed in a cloud computing infrastructure to evaluate personal driving assessment. The traffic signs detection realized by utilizing SIFT feature descriptor and the inter-vehicular distance estimation technique based on simple geometric constraints, are shown in detail.

Pedestrian Detection based on Gaussian Mixture ModelMultiresolution CoHOG

Recently, Co-occurrence histograms of oriented gradients (CoHOG) describes image features to calculate the co-occurrence of pixels allocated at the local level and has attracted attention as an effective object detection method. However, this method has some problems. For feature descriptions that focus on individual pixels, calculation cost and the number of dimensions tend to increase exponentially with respect to the number of pixels. Multiresolution CoHOG (MRCoHOG) can suppress such exponential increases to linear increase without reducing the classification accuracy. This paper proposes a procedure in which a feature plane is divided using a Gaussian mixture model and a histogram is automatically divided to establish a less costly method for performing MRCoHOG. Experimental results demonstrate that the proposed procedure is more effective than conventional procedures.

Simultaneous Estimation of Optical Flow and Its Boundaries based on the Dynamical System Model.

Optical flow is a velocity vector which represents the motion of objects in video images. Optical flow estimation is difficult in the neighborhood of flow boundary. To resolve this problem, Sasagawa (2014) proposes a modified dynamical system model in which one assumes that, in the neighborhood of flow boundaries, the brightness flows in the perpendicular direction, and considers the resulting corrections to the brightness constancy constraint. However, in that model, the correction is occurred even in place where the flow is continuous. We propose a new model, which switches the conventional model and the proposed model in Sasagawa (2014). As a result, we expect improvement of the estimate accuracy in place where the flow is continuous. We conduct numerical experiments to investigate the improvements that the proposed model yields in the estimation accuracy of optical flows.

Evaluation of Hardware Oriented MRCoHOG using Logic Simulation.

Human detection require high speed and high accuracy processing. One of the high performance techniques of the detection is multi-resolution co-occurrence histogram of oriented gradients (MRCoHOG). Since the calculation of co-occurrence requires a huge amount of processing resources, it is difficult to realize real-time human detection with MRCoHOG. Accordingly, hardware implementation is considered to be effective. In this paper, a hardware oriented MRCoHOG is proposed. In the proposed method, we simplify complicated calculation such as multiplications and square root operation for efficient hardware implementation. Experimental results show that the proposed method achieves better human detection rate than the ordinary method. Moreover, MRCoHOG is implemented in a digital circuit with the proposed method. According to logic simulation of the proposed circuit, the processing speed of the hardware implementation is 466 times higher than the software implementation.