Joint Recognition of the Moving and Stationary Objects in the Machine Vision Systems of Robots

Simultaneous detection of multiple stationary and moving obstacles in the near field of the mobile robots is a challenging task, since a robot has to detect a maximal possible number of obstacles, and ensure its movement without collisions. In this paper, the authors propose modified algorithms for detection of objects. Detectable objects are divided into two types: familiar objects (stationary obstacles, for example, a table, a chair, a computer, etc.), and unknown objects (moveable objects - people). The authors present specific recognition algorithms for each object type: the nearest neighbor search algorithm modified for the use with FLANN library and search trees (KNN) used for detection of the stationary obstacles; the built-in algorithms (Microsoft Kinect development kit-SDK) are intended for recognition of such movable objects as persons. The efficiency of the search algorithm of the nearest neighbors for detection of stationary objects is shown. This algorithm is implemented in FLANN library, which contains main algorithms for extraction of the handles of images and creation of indexes. The effectiveness of finding objects is increased due to application of SURF algorithm. Use of FLANN Library together with SURF algorithm satisfies the requirements for detection of objects in real time. The experimental results prove the effectiveness of the proposed approach in a vision system of a mobile robotic platform.

FLANN, SURF, мобильный робот, распознавание изображений, поиск ближайшего соседа, изображения глубины, предварительная обработка глубинных данных, Kinect SDK, FLANN, SURF, mobile robot, image recognition, nearest neighbor search, Kinect, depth image, deep data preprocessing, Kinect SDK

1. Navneet Dalai, Bill Triggs. Object Detection using Histograms of Oriented Gradients. URL: http://www.webcitation.org/6DvoEuAvL (дата обращения: 14.10.2014).

2. Ramisa A., Vasudevan S., Aldavert D., Toledo R., Lopez de Mantaras R. Evaluation of the SIFT Object Recognition Method in Mobile Robots. URL: http://www.iiia.csic.es/~mantaras/sift_eval.pdf (дата обращения: 14.10.2014).

3. Bay H., Ess A., Tuytelaars T., Van Gool L. Speeded-Up Robust Features (SURF) // Computer Vision and Image Understanding. 2008. Vol. 110, Issue 3. P. 346-359.

4. Marron R., Janet J. C# Implementation of SLAM Using the Microsoft Kinect. URL: http://old.ee.duke.edu/files/ece/GWD-D2012_Marron.pdf (дата обращения: 14.10.2014).

5. Shotton J., Fitzgibbon A. Cook M., Sharp Т., Finocchio M., Moore R., Kipman A., Blake A. Real-time human pose recognition in parts from single depth images // VISAPP International Conference on Computer Vision Theory and Applications. February 2013. Pp. 119-135.

6. Kefer M., Kubinger W. Evaluation of Kinect Depth Sensor for Use in Mobile Robotics // Annals of DAAAM for 2011 & Proceedings of the 22nd International DAAAM Symposium. 2011. Vol. 22, N. 1.

7. Knuth D. Section 6.1: Sequential Searching. Sorting and Searching. The Art of Computer Programming 3. Addison-Wesley. 1997. P. 396-408.

8. Muja M., Lowe D. G. Fast approximate nearest neighbors with automatic algorithm configuration // VISAPP International Conference on Computer Vision Theory and Applications. 2009. P. 331-340.

9. Mount D. M., Arya S. ANN: A Library for Approximate Nearest Neighbor Searching. URL: http://www.cs.umd.edu/~mount/ANN/ (дата обращения: 05.11.2014).

10. Connor M. The Simple, Thread-safe Approximate Nearest Neighbor (STANN). URL: https://sites.google.com/a/compgeom.com/stann/Home (дата обращения: 05.11.2014).

11. Forsyth D. A., Ponce J. Computer Vision: A Modern Approach // Robotics and Autonomous Systems. 2006. Vol. 11, N. 3. P. 84-92.

12. Обзор информации об алгоритме ближайшего соседа. URL: http://www.basegroup.ru/library/analysis/regression/knn/ (дата обращения: 05.11.2014).