|
Comparison of Characteristic Information in Gait Vibration |
|
Hiroto Ikeshita, Shunsuke Kido, Yuichi Nakamura(National Institute of Technology, Anan College, Japan), Yohei Saika(National Institute of Technology, Gunma College, Japan), Masahiro Nakagawa(Nagaoka University of Technology, Japan) |
| It is presumed that the personal characteristics information reflected the gait vibration data on floor. The gait vibration data are acquired, and one step, two steps and four steps patterns are extracted from the data. The Adaboost method is applied to pattern extraction from gait vibration data. Moreover, the personal characteristics information is analyzed from the one step, two steps and four steps patterns, and the possibility of personal identification is examined. The Principal Component Analysis is used to the characteristics evaluation of pattern for the personal identification. The p |
|
|
Mean-field Analysis of Bayesian Inference Using Expected A Posterior Estimation of Predicting Comfortable Environments Due to Air Conditioner |
|
Yohei Saika(Natl.Inst.Tech.GC, Japan), Masahiro Nakagawa(Nagaoka University of Technology, Japan) |
| On the basis of the mean-field theory established in statistical physics we investigate the Bayesian inference using the the expected a posterior (EAP) estimation for predicting a set of environmental variables which realize comfortable environments due to air conditioner. In this method, the posterior probability is estimated using the model prior which realizes the thermal comfort via thetemperature-humidity index around the optimal value and the likelihood which expresses expressing transition probability from each comfortable state to realistic one. Then, using the mean-field theory, we find |
|
|
Performance Evaluation of a Beach Cleaning Robot "Hirottaro 3" in an Actual Working Environment |
|
Tomoyasu Ichimura(National Institute of Technology, Gunma College, Japan), Shin-ichi Nakajima(Department of Mechanical and Control Engineering, Japan) |
| We have been developing a compact beach cleaning robot "Hirottaro 3". In order to collect small refuse effectively on a sandy surface, the robot was equipped with a mechanism that collected items as if humans clean a floor with a broom and dustpan. Furthermore, the robot was capable of traveling autonomously on the sandy beach that had insufficient natural landmarks by self-localization using poles and a scanning range finder. This paper reports the performance evaluation of refuse collection and autonomous navigation on the sandy beach. |
|
|
Covariance Matrix of a Probability Distribution for Image Dictionaries in Compressed Sensing |
|
Toshiaki Aida(Okayama University, Japan) |
| Sparse representation is one of the principles for the most effective signal processing. The framework of signal processing based on it is called compressed sensing.
In our previous work, we successfully derived an analytical expression of the probability distribution followed by image dictionaries for the images generated by the Gaussian model. However, we have found that it has a difficulty of a divergent covariance matrix, which is needed for an analytical performance evaluation of image processing by compressed sensing. Therefore, it is the purpose of this paper to solve the difficulty. |
|
|
Study of Measurement Method in Inter-Vehicle Distance Using Hu Moment Invariants |
|
Nobuo Sasaki, Takeshi Funatsu(National Institute of Technology, Gunma College, Japan) |
| This paper proposed a method to detect the license plate by using the Hu moment invariants. Adaptive thresholding process made the detection robust against changes of brightness caused by weather changes. Adaptive thresholding process achieved the detection rate of 93.2%. This result shows that the robustness of the system has
been improved. At a distance of 2 to 15 m, the error rate was about 4 to 6%, and the average absolute error rate was 4.75%. The average time for all processes was 44.67 ms, which shows that this method operates at a maximum speed of about 22.39 fps. |
|
|
Recognizing Human-Object Interactions via Target Localization |
|
Sunyoung Cho, Jihun Park, Young Sook Shin, Sang-ho Lee(Agency for Defense Development, Korea) |
| We propose a target prediction model that aims to identify regions relevant to the human-object interactions. Our model predicts the precise target location relating to the specific action by formulating it to FCN that enables fine-grained localization. We jointly learn the appearance and location of the target by exploiting the target-specific segmentation information. We show that our target prediction model outperforms state-of-the-art methods in identifying small and occluded objects, and its result can be used to improve the recognition of human-object interactions. |
|