Optik akışın hesaplanması ve yapay sinir ağları ile yorumlanarak mobil robotlar için engel tespiti ve kaçınma davranışında kullanılması

Görüntü sensörleri bilgisayar teknolojisinin hızlı bir şekilde gelişmesiyle robotik araştırmalarda yoğun bir şekilde kullanılmaya başlanmıştır. Çok değişik uygulamalardan birisi de görüntünün Optik Akışı üzerinde çalışmak suretiyle mobil robotun navigasyon yaptığı ortam hakkında bilgi toplamaktır. Optik akışı görüş alanında nesnelerin hareketi olarak düşünebiliriz. Ortamda bağıl bir hareket söz konusu ise ve stereo görüntü alınabiliyorsa, elde edilecek bilgiler mobil robotun navigasyon ortamında engel tespiti ve engellerden kaçınma davranışları için kullanılabilmektedir. Optik akış, üzerinde çok uzun süredir çalışılan bir konudur. Ancak bütün görüntü tabanlı uygulamalarda olduğu gibi hesaplama yükünden dolayı gerçek zamanlı çalışmalarda kullanılmasında pek çok zorluklarla karşılaşılmaktaydı. Son yıllarda yapılan çalışmalar ile optik akışın hesaplanmasına yönelik pratik teknikler ortaya konmuştur. Bu çalışmada optik akıştan ve Yapay Sinir Ağlarından faydalanılmak suretiyle mobil robot için engel tespiti ve engellerden sakınma davranışı ortaya konmaya çalışılmıştır. Çalışmalar Matlab simülasyon ortamında gerçekleştirilen deneysel sonuçlarla desteklenmiştir. Gerçek ortamdan alınan görüntülerden, optik akışları hesaplanmak suretiyle oluşturulan bir veri seti ile,yine matlab ortamında oluşturulan çok katmanlı bir perseptron YSA' nı eğitmek suretiyle engel tanıyacak bir sistem ortaya konulmuştur. YSA eğitiminde ise Levenberg- Marquardt Öğrenme Algoritması kullanılmıştır. Elde edilen deneysel sonuçlar, bu metodolojinin gerçek zamanlı olarak uygulanabileceği konusundaki düşünceleri kuvvetlendirmektedir.

The calculation of optical flow and interpretation of the results using artificial neural network in order to use for the obstacle detection and avoidance behaviors of the mobile robots

With the rapid improvement of computer technology, Visual-based sensors have gained an intense popularity and consequently have begun to be utilized extensively in robotic research. Among the various applications in robotics, one of the most popular concepts is gathering information from the navigation environments for mobile robots by working on optical flow of vision which is derived from a stereo camera located on the robots. We can determine from the optical flow the movement of the objects within the area of robotic vision. If a relative motion in the environment, whether from objects or the mobile robot, is present, then the information that can be gathered from this environment is enough for the mobile robot to execute its obstacle detection and avoidance behaviors. Optical Flow is a concept which has been worked on for quite a long time. But due to problems which prevail on all visual based applications, such as computing difficulties and slow rate of getting results, researchers have come across with so many difficulties that deter them from use in real time applications, especially in robotics. But as the latest research and techniques have come to view, new practical methods were put forward. In this study, by making use of optical flow calculation and multi layer perceptron Artificial Neural Network, a methodology has been tried to be put forward for mobile robot obstacle detection and avoidance behavior. The study of methodology has been supported by experimental results that were obtained from Matlab simulation environments. The images of the views were taken from the real navigation environment and then optical flow calculations for all images were obtained via matlab simulink blocks that were created in advance, as an algorithm which can calculate optical flows from stereo visions. As optical flows of each pair of stereo views were derived, a data base was constituted in order to train the multi layer perceptron. By the help of the data set and the Levenberg- Marquardt learning algorithm, a neural network which was well trained in Matlab environment in order to detect the presence of obstacles was created. Experimental results, obtained during the study have strengthened the ideas which have supported the usage of the Optical Flow via Artificial Neural Network in mobile robotics for obstacle detection and avoidance behaviors.

___

  • [1] Josh Clanton, 2005. Optical Flow Computation Using the Horn &Schunck Algorithm.
  • [2] Kahlouche Souhila and Achour Karim,2005. Optical Flow based robot obstacle avoidance.
  • [3] Jorg Rett and Jorge Dias, 2007. Autonomous Robot Navigation-A study using Optical Flow and log-polar image representation.
  • [4] Toby Low and Gordon Wyeth,2005. Obstacle Detection using Optical Flow. School of Information Technology and Electrical Engineering University of Queensland.
  • [5] Jun Li, 2006.Learning Reactive Behaviors with Constructive Neural Networks in Mobile Robotics.Yüksek Lisans Tezi.
  • [6] Daniel Oberhoff, Andy Stynen, and Marina Kolesnik, 2006.Neural Network Architecture for Modeling the Joint Visual Perception of Orientation, Motion,and Depth.
  • [7] Philippe Künzle, 2005. Building Topological Maps For Robot Navigation Using Neural Network. Department of Computer Science McGill University, Montréal. Doktora Tezi.
  • [8] Kahlouche Souhila and Achour Karim, 2006. Optical Flow based robot obstacle avoidance.
  • [9] Horn, K.P. & Schunck, B.G.,1981. Determining optical flow. Artificial intelligence.
  • [10] Eliete Maria de Oliveira Caldeira,Hans Jörg Andreas Schneebeli,Mário Sarcinelli-Filho,2005.An optical flow-based sensing system for reactive mobile.
  • [11] Thorsten Gernoth, 2004.Sequential Feature Matching and Structure from Motion for Robot Navigation.Göteborg University and Chalmers University of Technology.Master Tezi
  • [12] Sameer Agarwal, Satya P. Mallick, David Kriegman, and Serge Belongie,2006.On Refractive Optical Flow.University of California, San Diego, La Jolla CA 92093, USA.
  • [13] J. L. Barron, D. J. Fleet, and S. S. Beauchemin, 1994. Performance of optical flow techniques. International Journal of Computer Vision,
  • [14] Jason Campbell, Rahul Sukthankar, Illah Nourbakhsh ,2004. Techniques for Evaluating Optical Flow for Visual Odometry in Ekstreme Terrain. Proceedings of IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2004.