Development of Mobile Robot System Equipped with Camera and Laser Range Finder Realizing HOG-Based Person Following and Autonomous Returning

Transcription

1 Awai, M. et al. Paper: Development of Mobile Robot System Equippe with Camera an Laser Range Finer Realizing HOG-Base Person Following an Autonomous Returning Masashi Awai, Atsushi Yamashita, Takahito Shimizu, Toru Kaneko, Yuichi Kobayashi,anHajimeAsama Department of Mechanical Engineering, Shizuoka University Johoku, Naka-ku, Hamamatsu-shi, Shizuoka , Japan Department of Precision Engineering, The University of Tokyo Hongo, Bunkyo-ku, Tokyo , Japan [Receive September 10, 2013; accepte November 7, 2013] In this paper, we propose a mobile robot system which has functions of person following an autonomous returning. The robot realizes these functions by analyzing information obtaine with camera an laser range finer. Person following is performe by using HOG features, color information, an pattern of range ata. Along with person following, a map of the ambient environment is generate from range ata. Autonomous returning to the starting point is performe by applying potential metho to the generate map. We verifie the propose metho by experiment using a wheel mobile robot in an inoor environment. Keywors: mobile robot, laser range finer, camera, person following 1. Introuction In recent years, introuction of autonomous mobile robots to environments close to us is expecte. Examples inclue shopping cart robots returning automatically to the shopping cart she after shopping an guie robots irecting the way from the current location to the starting point in unknown environments. A robot that accomplishes these purposes nees functions of person following an autonomously returning to the starting point [1 3]. In reference [1] an reference [2], returning was performe after arrival at the estination by manual control. However, manual control of the robot requires great care. In reference [3], person following an autonomous returning were performe. However, autonomous returning was performe only by tracing the route recore on the outwar way. We previously propose in reference [4] a mobile robot system which has functions of person following an autonomous returning to the starting point while avoiing obstacles. In this system, person following was performe by etection of moving object using Laser Range Finer (LRF) an person extraction using color information. However, etection of moving object was unstable epening on the walking spee. Therefore, We nee more stable person following metho. Person following methos were propose in references [5 7]. A metho in reference [5] etecte persons by template matching of legs-like pattern for LRF ata. To follow the target person, it employe particle filtering whose initial particles were generate from infrare image. A metho in reference [6] use stereo camera an etecte persons by epth templates of person shape. It employe the Extene Kalman filter for person following. A metho in reference [7] use LRF an omni-view camera. LRF ata were use for etecting caniate of person regions an omni-view camera images were use for extracting person regions base on HOG (Histogram of Oriente Graients) features. It also employe particle filtering for person following. The abovementione methos for person following use one or two criteria for person etection. Here, note that aition of criterion for person etection increases the robustness of person following if one criterion itself works well. Therefore, in this paper we propose a mobile robot which has a function of robust person following realize by integration of three criteria for person etection base on pattern of LRF ata, HOG features an color histogram. The robot also has a function of self localization for map generation on the outwar way, an a function of autonomous return with obstacle avoiance. 2. Outline In this paper, we verify the system using a mobile robot equippe with a Laser Range Finer (LRF) an a camera (Fig. 1). The mobile robot acquires two imensional (2-D) range ata of 180 forwar by the LRF. The mobile robot also acquires the image in the front irection by the camera. 68 Journal of Robotics an Mechatronics Vol.26 No.1, 2014

2 Development of Mobile Robot System with Camera an LRF Camera LRF Fig. 3. Points of ata evaluation. Fig. 1. Mobile robot. Fig. 2. Environment. The mobile robot etects an follows a person by using the LRF an the camera when moving on the outwar way. At the same time, the mobile robot generates a map with range ata measure by the LRF. The mobile robot generates a potential fiel from the generate map by an artificial potential metho. Then, it moves on the return way along graient irections of the generate potential fiel. At the same time, the mobile robot avois obstacles not recore in the map by reconstruction of the potential fiel. An operating environment of the mobile robot is a flat floor where the mobile robot moves in 2-D space (Fig. 2). In the operating environment, multiple objects an peestrians usually exist. In this situation, person following by a single metho of person etection may be ifficult. It will be realize by combining multiple methos of person etection. 3. Person Following on Outwar Way The mobile robot performs person following an map generation on the outwar way. In this paper, the mobile robot follows the person by using person etection. Person etection is performe by evaluating a value of person likelihoo. The value of person likelihoo can be evaluate on the pattern of range ata an the HOG features in acquire images. However, this evaluation is ineffective to follow the person in environment where more than one person exist. The reason for this is that the pattern of LRF ata an HOG features are not specific to the person who shoul be followe but common to all per- sons. Uner an assumption that peestrians o not wear clothes with similar color to the person to be followe, color information is ae to the pattern of range ata an HOG features. Thus, person following is performe by using particle filter base on pattern of range ata, HOG features an color information Tracking by Particle Filter We use a particle filter [8] for person following. The particle filter estimates the position of the person to be followe on a horizontal plane. In this stuy, ensity of particle istribution represents the probability of existence of the person to be followe. By calculating particle positions accoring to the particle filter algorithm, position of the person to be followe can be estimate. In particle filter algorithm, a particle is assigne a weight at each particle position. However, calculation of weight at each particle position is computationally expensive. Thus, assigning the weight to the particle is performe at representative points. Representative points are set at the positions of range ata basically. But here we have to notice that points in between legs are not evaluate through the process. Therefore, process to cover points in between legs shoul be performe. Fig. 3 shows process of etermining representative points. First, the robot acquires range ata. Distance D(θ) an angle θ are obtaine as the range ata in polar coorinates. Initially, representative points R = (U(θ)cos(θ),U(θ)sin(θ)) are set using the range ata as U(θ)=D(θ). Process to cover points in between legs is performe by Eq. (1), which will be applie to each range ata. U(θ) = min D(θ + l) k<l<k if (D(θ + l + k) < D(θ + l)),.... (1) ( ) W k = arctan, (2) 2D(θ + l) where W is estimate value of length in between legs. Next, values of person likelihoo are evaluate at each representative position. In the particle filter algorithm, particles near to representative points are assigne a weight as evaluate values V( R). Thus, calculating V( R) is performe only at representative points close to particles. Particle filter is performe as follows. (i) Initial particles are istribute aroun the person with ranom noise. Journal of Robotics an Mechatronics Vol.26 No.1,

3 Awai, M. et al. D(θ) [m] [egree] θ Fig. 5. Acquire image. Fig. 4. Range ata. T 1 ( k) T 2 ( k) T 3 ( k) (ii) Particles are move base on the system moel given by Eq. (3). x t t 1 = x t 1 + v t 1 T, (3) 0 k 0 k 0 k (a) Close legs (b) Open legs (c) Single leg where x t is an estimate position vector of the person on a horizontal plane at time t, v t is velocity vector of the person at time t, an T is time of upating cycle. v t is estimate by the subtraction of x t 1 an x t 2. T 4 ( k) 0 k T 5( k ) 0 k 0 T 6 ( k ) k (iii) Each particle is assigne a weight w as following equation. ) V(R)exp ( D2 2σ w =, (4) 2πσ where σ is position error. Position error inclues error of self-location estimation an measuring error. However, these error are har to efine. Thus, position error is efine as the constant value empirically. D is istance between the particle an the nearest range ata, V(R) is value of likelihoo of the nearest representative point. (iv) Estimate position x t is etermine by calculating center of gravity of particles. Position x t is regare as the position of the person to be followe at time t. Thus, the mobile robot move towar the position x t at time t. (v) New particles are selecte in escening orer accoring to the weight as resampling process. First, process (i) is performe. Next, processes (ii), (iii) an (iv) are performe. After that, processes (ii), (iii), (iv) an (v) are performe repeately Evaluation on Pattern of Range Data Person etection is performe by using pattern of range ata. Range ata is acquire as shown in Fig. 4 in the environment (Fig. 5). In the reference [5], person following was performe by template matching from LRF ata. In the stuy, person following was achieve by using templates with 7 position-relation patterns of person () Right oblique legs (open) (e) Left oblique legs (open) (f) Right oblieque legs (close) 0 T 7 ( k) (g) Left oblieque legs (close) k Fig. 6. Leg templates. legs. Therefore, as the evaluation on pattern of range ata, pattern matching is performe by using range ata an 7 templates of legs that are the same as [5] in our stuy. The 7 patterns are shown in Fig. 6, where relative position of legs are varie as close, open, left forwar, right forwar an so on. While following the person, matching scores are obtaine by performing pattern matching at all representative positions R(θ,U(θ)). Matching score L( R) is obtaine at each R by Eq. (5): L( R)= S( R), (5) S( R)= min A 1 i N k= A ( W A = arctan 2U(θ) (D(θ + k) U(θ)) T i (k), (6) ), (7) where N is number of template, A is with of templates, is the value set to a maximum value of ((D(θ + k) U(θ))). 70 Journal of Robotics an Mechatronics Vol.26 No.1, 2014

4 Development of Mobile Robot System with Camera an LRF Person Non-person Frequency Fig. 7. Dataset Evaluation on HOG Features Person etection in acquire images is performe by using Real AaBoost algorithm [9] with HOG features. The valiity of HOG features for person etection is verifie in the reference [10]. Thus, classifier of Real AaBoost for person etection is generate from HOG features of sample images of person an backgroun. Daimler Peestrian Classification Benchmark Dataset [a] is use as sample images (Fig. 7). Real AaBoost constructs the strong classifier as linear combination of weak classifier. H = D =1 h, (8) where D is the number of weak classifier, H is value of strong classifier, h is value of weak classifier. We use value of H as the constructe strong classifier value of Real AaBoost. For person etection on HOG features, image regions to perform person etection are selecte in acquire images. H( R) is acquire at each representative point R. By setting the size of person, size an position of the image region are etermine accoring to each representative point R Evaluation on Color Histogram Detecting the person who shoul be followe is performe by using color histogram. The color histogram of hue h an saturation s is mae (an example of huesaturation histogram is epicte in Fig. 8 with an image of a person in an inoor environment). Number of bins of hue saturation set to This setting is base on experiment with human ientification conucte by Takahashi et al. [11], where histogram of h = 16 s = 8was verifie to be effective to make ientification robust to the changes of brightness. The mobile robot acquires color information on the person to be followe by the mobile robot before the person following begins. While the mobile robot follows the person, the color information is acquire from the image region which is selecte accoring to each position of representative point. Assuming that the orinary size of person an the istance between camera an person are known, image regions can be etermine accoring to the representative point R. Then, the color histogram is mae an the similarity with the color histogram mae before the person following begins is calculate Saturation Hue Fig. 8. An example of color histogram. The Bhattacharyya coefficient [12] is use for calculating the similarity with the histograms. Eq. (9) shows Bhattacharyya coefficient C( R) of point R. C( R)= h b h=0 s b s=0 H t (h,s) H(h,s, R),... (9) where H t (h,s) is the frequency of each bin of the color histogram of the person that is acquire before the mobile robot begins person following. H(h,s, R) is the frequency of each bins of the color histogram acquire in image region for position R while the mobile robot follows the person. h b is the number of bins of hue h, ans b is the number of bins of saturation s Integration of Evaluate Values by PCA The integrate value of person likelihoo V( R) is calculate by applying PCA (Principal Component Analysis) to L( R), H( R) an C( R). Before PCA is performe, averaging proceure is performe. This proceure is introuce because large ifferences of values among representative points in the near istance can be noise for PCA. Average values L s ( R),H s ( R),C s ( R) are calculate by following equations. L s ( R i )= H s ( R i )= C s ( R i )= a i j a i j a i j L( R j ), (10) H( R j ), (11) C( R j ), (12) where R j is representative point having the istance between R i an R j not more than threshol a, a i is number of R j. Next, for processing PCA, stanarize value (V L (R), V H (R), V C (R)) is calculate from L s ( R), H s ( R) an C s ( R) Journal of Robotics an Mechatronics Vol.26 No.1,

5 Awai, M. et al. as the followings. V L ( R)= L s( R) L s ( R) s L, (13) : Graient irection : Configuration obstacle Obstacle Map ata V H ( R)= H s( R) H s ( R) s H, (14) V C ( R)= C s( R) C s ( R) s C, (15) where L( R),H( R),C( R) are averages, s L,s H,s C are variances. PCA is processe by calculating eigenvector from variance-covariance matrix from V L ( R), V H ( R), V C ( R). S(V L,V L ) S(V L,V H ) S(V L,V C ) = S(V H,V L ) S(V H,V H ) S(V H,V C ), (16) S(V C,V L ) S(V C,V H ) S(V C,V C ) where S(V i,v j ) is covariance of V i an V j. Finally, the integrate value of person likelihoo V( R) is calculate by using the primary ingreient (α,β,γ)an eigenvalue of the primary ingreient λ. V( R)=αV L ( R)+βV H ( R)+γV C ( R)+2 λ. (17) To avoi negative effect by a large noise, a coefficient is set zero when its corresponing component has negative value. Initial path (a) Initial path. Recalculation path (b) Recalculation path. Fig. 9. Reconstruction of potential fiel. Then the robot moves on the return way along a graient irection of the generate potential fiel. For the robot to avoi obstacles not recore in the map, the LRF measures an ambient environment while the robot moves on the return way an the robot reconstructs a potential fiel. Fig. 9 shows reconstruction of a potential fiel. If the robot etects an obstacle which i not exist on the outwar way, the obstacle is ae to the map ata mae on the outwar way (Fig. 9(a)). Then the robot reconstructs a potential fiel to avoi obstacles (Fig. 9(b)). This makes the movement of the robot safe on the return way. 4. Map Generation on Outwar Way The mobile robot generates the map of ambient environment while it moves on the outwar way. The LRF is use to measure the ambient environment uring the mobile robot movement, an the ambient environment map is generate by integrating each measurement ata. Measurement ata integration nees an accurate self-location estimation of the mobile robot. In this stuy, the estimation is mae by ea reckoning. However, ea reckoning has a problem of error accumulation cause by wheel slipping. In orer to ecrease this error accumulation, the robot aligns each measurement ata by the ICP algorithm [13]. Moving objects o not exist in the same place. Therefore, it is necessary to remove moving objects from the map. The mobile robot removes moving objects by a metho in the reference [14]. To remove a moving object, subtraction of acquire range ata at ifferent robot positions is performe. The map of static object can be acquire by using this metho. 5. Motion on Return Way The mobile robot moves on the return way accoring to the Laplace potential metho [15]. The robot generates the potential fiel in the map obtaine on the outwar way. 6. Experiment 6.1. Experimental Device We use the mobile robot Pioneer 3 of MobileRobots, Inc. (Fig. 2). The robot has 2 rive wheels an 1 caster. It s maximum spee is 400 mm/sec. It turns with the velocity ifferential of right an left wheels. The LRF is moel LMS by SICK. It is equippe at a height of 300 mm above the groun. The sensing range is 180 in one plane an the resolution is 0.5. The camera is moel C910 by logicool. Its horizontal angle of view is about 70. The camera is equippe at a height of 800 mm above the groun. As the specs on computers, CPU is Intel Core 2Duo T GHz, an memory is 3.5 GB Experiment Environment We conucte experiment in which the mobile robot follows a person to the target point an then returns to the starting point. Experiment environment was a corrior with a flat floor. Fig. 2 shows the experiment environment. The with of corrior is about 2 m. Wall materials are metal an concrete. There were two peestrians on vinyl flooring in experiment environment. We set position error σ = 50 mm, parameter of LRF etection = 700 mm, parameters of color histogram h b = 16, s b = 8, an parameter a = 50 mm. 72 Journal of Robotics an Mechatronics Vol.26 No.1, 2014

6 Development of Mobile Robot System with Camera an LRF V L θ [egree] Fig. 10. Acquire image (A). Fig. 13. Evaluation value of LRF V L (θ) (A). Fig. 14. Person etection by HOG features (A). V H Fig D map image (A). θ [egree] Fig. 15. Evaluation value of HOG V H (θ) (A). Fig. 12. Particles in 2-D map image (A). V C 6.3. Experimental Result On the outwar way, the mobile robot followe the person. Fig. 10 shows an image acquire with the camera while the robot followe the person. Figure 11 shows particles an a 2-D map image generate by range ata acquire with the LRF. Gray points in 2-D map are particles. Fig. 12 shows the magnifie map aroun particles in Fig. 11. Figure 13 shows evaluation value on shape of range ata V L (θ). The horizontal axis in Fig. 13 inicates the view angle from the robot (the positive an negative values correspon to the right an left angle, respectively). In Fig. 13, it is shown that high values appear in the vicinity of the angle where the person exists. Figure 14 shows person etection by HOG features. Fig. 15 shows evaluation value on HOG features V H (θ). In Fig. 14, white rectangles are the regions whose value of V H (θ) exceee the threshol. It is shown that high values appear in the vicinity of the angle where the person exists. θ [egree] Fig. 16. Evaluation value of color histogram V C (θ) (A). Figure 16 shows evaluation value on color information V C (θ). The obstacle which has the similar color to the person to be followe was existing in the right sie of the person. It is shown that high values appear in the vicinity of the angle where the obstacle exists (vicinity of 13 ). Finally, Fig. 17 shows the integrate value of person likelihoo V(θ). It is shown that the highest values appear in the vicinity of the angle where the person to be followe exists. Journal of Robotics an Mechatronics Vol.26 No.1,

7 Awai, M. et al. V e V L θ [egree] θ [egree] Fig. 17. Integrate evaluation value V(θ) (A). Fig. 21. Evaluation value of LRF V L (θ) (B). Fig. 18. Acquire image (B). Fig. 22. Person etection by HOG features (B). V H θ [egree] Fig. 23. Evaluation value of HOG V H (θ) (B). Fig D map image (B). V C θ [egree] Fig. 24. Evaluation value of color histogram V C (θ) (B). Fig. 20. Particles in 2-D map image (B). V Figure 18 shows an image acquire with the camera while the peestrian passe. In Fig. 18, the right person is to be followe by the mobile robot. Figs. 19 an 20 show particles an a 2-D map image. Fig. 21 shows V L (θ). Fig. 22 shows person etection by HOG features. Fig. 23 shows V H (θ). Fig. 24 shows V C (θ). Fig. 25 shows V(θ). θ [egree] Fig. 25. Integrate evaluation value V(θ) (B). 74 Journal of Robotics an Mechatronics Vol.26 No.1, 2014

8 Development of Mobile Robot System with Camera an LRF Table 1. Ratio of frame whose person s point has highest value. klrf HOG Color Integrate (V L ) (V H ) (V C ) (V ) Number of frame Rate of frame [%] e ) ( m LRF ata Robot trajectory Goal Start e e e e (m) Fig. 26. Generate map an trajectory of mobile robot on the outwar way. e ) (m LRF ata Robot trajectory Goal Start e e e e New obstacle (m) Fig. 27. Generate map an trajectory of mobile robot on the return way. In Figs. 21 an 23, it is shown that high values appear in the vicinity of the angle where the person exist (vicinities of 5 an 20 ). In Fig. 24, it is shown that high values appear only in the vicinity of the angle where the person to be followe exists. Finally, in Fig. 25, it is shown that the highest values appear in the vicinity of the angle where the right person to be followe exists. Thus, it is more effective to use integrate value than to use LRF value or HOG value. The number of acquire images an range ata uring person following was 349 frame. The number of frames containing objects outsie the person to be followe was 247 frame. Table 1 shows ratio of the frame which the highest value appears where the person to be followe exists among frames containing objects outsie the person to be followe. It is shown that integrate value has the highest ratio. Figure 26 shows the generate map an the robot trajectory on the outwar way. It is shown that stationary object map was generate by moving object etection. Figure 27 shows the trajectory of the robot on the return way. The robot move on the return way by using the map which ha been generate on the outwar way. On the return way, the obstacle that i not exist on the outwar way appeare. It is shown that the robot returne to the starting position while avoiing the obstacle. These results show that the mobile robot can etect an follow the person by the propose metho in the experimental environment an can return to the starting point while avoiing obstacles. The error between the starting point in Fig. 26 an the goal point in Fig. 27 maybecausebyerroroficp metho. The reason for this is that features along the moving irection in Figs. 26 an 27 foralignmentoficp metho were sparse in experimental environment. Journal of Robotics an Mechatronics Vol.26 No.1,

9 Awai, M. et al. 7. Conclusion In this paper, we constructe the mobile robot system that has functions of person following an returning to the starting location autonomously while avoiing obstacles. Person following was achieve by using pattern of range ata, HOG features an color information. Map generation was achieve by the ICP algorithm an the moving object etection. The robot returne to the starting point accoring to the Laplace potential metho with generate map an a path of avoiing obstacle is generate by reconstructing a potential fiel. As future works, we nee to conuct experiments in more iverse situations (many people, longer istance). This requires functions of supporting occlusion an more precise localization in ynamic environment. Furthermore, in the situation that peestrians wearing same color clothes exist, the propose metho may not be effective. In such cases, utilization of time-series information will play more important role. More precise an robust system moel in the particle filter (or other filter techniques) will be require. Acknowlegements This work was in partly supporte by the Kayamori Founation of Informational Science Avancement, an MEXT KAKENHI, Grantin-Ai for Young Scientist (A), [12] T. Kailath, The Divergence an Bhattacharyya Distance Measures in Signal Selection, IEEE Trans. on Communication Technology, Vol.COM-15, No.1, pp , [13] P. J. Besl an N. D. Mckay, A Metho for Registration of 3-D Shapes, IEEE Trans. on Pattern Analysis an Machine Intelligence, Vol.14, No.2, pp , [14] S. Iwashina, A. Yamashita, an T. Kaneko, 3-D Map Builing in Dynamic Environments by a Mobile Robot Equippe with Two Laser Range Finers, Proc. of the 3r Asia Int. Symposium on Mechatronics, TP1-3(1), pp. 1-5, [15] K. Sato, Dealock-free Motion Planning Using the Laplace Potential Fiel, Proc. of the Joint Int. Conf. on Mathematical Methos an Supercomputing in Nuclear Applications, pp , Supporting Online Materials: [a] Daimler Peestrian Classification Benchmark Dataset, Masashi Awai Ctec, Inc. Aress: 35th Floor, Mori Tower, Roppongi Hills, Roppongi, Minato-ku, Tokyo, Japan 2013 Receive Master egree from Grauate School of Engineering, Shizuoka University Ctec, Inc. References: [1] M. Misawa, T. Yoshia, an S. Yuta, A Smart Hancart with Autonomous Returning Function, J. of Robotics Society of Japan, Vol.25, No.8, pp , 2007 (in Japanese). [2] T. Lixin an S. Yuta, Mobile Robot Playback Navigation Base on Robot Pose Calculation Using Memorize Omniirectional Images, J. of Robotics an Mechatronics, Vol.14, No.4, pp , [3] N. Tsua, S. Harimoto, T. Saitoh, an R. Konishi, Mobile Robot with Following an Returning Moe, Proc. of the 18th IEEE Int. Symposium on Robot an Human Interactive Communication (RO- MAN2009), ThB1.3, pp , [4] T. Shimizu, M. Awai, A. Yamashita, an T. Kaneko, Mobile Robot System Realizing Human Following an Autonomous Returning Using Laser Range Finer an Camera, Proc. of the 18th Korea- Japan Joint Workshop on Frontiers of Computer Vision (FCV2012), [5] S. Okusako an S. Sakane, Human Tracking with a Mobile Robot using a Laser Range-Finer, J. of Robotics Society of Japan, Vol.24, No.5, pp , 2006 (in Japanese). [6] J. Satake an J. Miura, Person Following of a Mobile Robot using Stereo Vision, J. of Robotics Society of Japan, Vol.28, No.9, pp , 2010 (in Japanese). [7] M. Saito, K. Yamazaki, N. Hatao, R. Hanai, K. Okaa, an M. Inaba, Peestrian Detection using a LRF an a Small Omni-view Camera for Outoor Personal Mobility Robot, Proc. of the 2010 IEEE Int. Conf. on Robotics an Biomimetics (ROBIO 2010), pp , [8] M. Isar an A. Blake, Conensation-conitional ensity propagation for visual tracking, Int. J. of Computer Vision, Vol.29, No.1, pp. 5-28, [9] R. E. Schapire an Y. Singer, Improve boosting algorithms using confience rate preictions, Machine Learning, No.37, pp , [10] N. Dalal an B. Triggs, Histograms of Oriente Graients for Human Detection, Proc. of the 2005 IEEE Computer Society Conf. on Computer Vision an Pattern Recognition (CVPR2005), pp , [11] H. Takahashi, K. Nakamura, H. Zhao, an R. Shibasaki, Human Ientification Using Laser Scanners an Image Sensors, Proc. of Asian Conf. on Remote Sensing 2007, TS24-2, Atsushi Yamashita Associate Professor, Department of Precision Engineering, The University of Tokyo Aress: Hongo, Bunkyo-ku, Tokyo , Japan Assistant Professor, Shizuoka University Visiting Associate, California Institute of Technology Associate Professor, Shizuoka University Associate Professor, The University of Tokyo Main Works: Motion Planning of Multiple Mobile Robots for Cooperative Manipulation an Transportation, IEEE Trans. on Robotics an Automation, Vol.19, No.2, pp , Apr Parallel Line-base Structure from Motion by Using Omniirectional Camera in Texture-less Scene, Avance Robotics, Vol.27, No.1, pp , Jan Membership in Acaemic Societies: The Institute of Electrical an Electronics Engineers (IEEE) Association for Computing Machinery (ACM) The Robotics Society of Japan (RSJ) The Japan Society of Mechanical Engineers (JSME) The Japan Society for Precision Engineering (JSPE) The Institute of Electronics, Information an Communication Engineers (IEICE) The Institute of Electrical Engineers of Japan (IEEJ) Information Processing Society of Japan (IPSJ) The Institute of Image Information an Television Engineers (ITE) The Society of Instrument an Control Engineers (SICE) Society for Serviceology 76 Journal of Robotics an Mechatronics Vol.26 No.1, 2014

10 Development of Mobile Robot System with Camera an LRF Takahito Shimizu Iwata Shinkin Bank Yuichi Kobayashi Associate Professor, Grauate School of Engineering, Shizuoka University Aress: Nakaizumi, Iwata-shi, Shizuoka, Japan 2012 Receive Master egree from Grauate School of Engineering, Shizuoka University Iwata Shinkin Bank Toru Kaneko Professor, Grauate School of Engineering, Shizuoka University Aress: Johoku, Naka-ku, Hamamatsu-shi, Shizuoka, Japan Research Scientist, RIKEN Bio-mimetic Control Research Center Associate Professor, Tokyo University of Agriculture an Technology Associate Professor, Shizuoka University Main Works: Planning-Space Shift Motion Generation: Variable-space Motion Planning Towar Flexible Extension of Boy Schema, J. of Intelligent an Robotic Systems, Vol.62, Issue 3-4, Jun Membership in Acaemic Societies: The Institute of Electrical an Electronics Engineers (IEEE) The Robotics Society of Japan (RSJ) The Society of Instrument an Control Engineers (SICE) The Japan Society of Mechanical Engineers (JSME) The Japan Society for Precision Engineering (JSPE) Aress: Johoku, Naka-ku, Hamamatsu-shi, Shizuoka, Japan Nippon Telegraph an Telephone Corporation NTT Human Interface Laboratories Faculty of Engineering, Shizuoka University Main Works: Chroma Key Using a Checker Pattern Backgroun, IEICE Trans. on Information an Systems, Vol.90-D, No.1, pp , Jan Membership in Acaemic Societies: The Institute of Electrical an Electronics Engineers (IEEE) The Institute of Electronics, Information an Communication Engineers (IEICE) The Institute of Image Information an Television Engineers (ITE) Hajime Asama Professor, Department of Precision Engineering, The University of Tokyo Aress: Hongo, Bunkyo-ku, Tokyo , Japan Research Associate, RIKEN (The Institute of Physical an Chemical Research) Professor, RACE (Research into Artifacts, Center for Engineering), The University of Tokyo Professor, Department of Precision Engineering, School of Engineering, The University of Tokyo Main Works: Aaptive Division-of-Labor Control Algorithm for Multi-robot Systems, J. of Robotics & Mechatronics, Vol.22, No.4, pp , Aug Membership in Acaemic Societies: The Institute of Electrical an Electronics Engineers (IEEE) The Robotics Society of Japan (RSJ) The Japan Society of Mechanical Engineers (JSME) The Japanese Society of Instrumentation an Control Engineers (SICE) Journal of Robotics an Mechatronics Vol.26 No.1,