Omnidirecional image database of virtual and real environment for mobile robot localization

We used a small mobile robot adapted to locomotion in a home environment, see Fig. 4(c). It consisted of a metal platform equipped with engines, logic boards, batteries, and a system of interconnected gears with two pairs of wheels. The robot has its own software that communicates with a microcomputer controlled by telemetry. A camera along with the Dot, a lens that shoots 360 degree images was used.

figOmni

Fig. 4. Topological map of the test environment: a) top; b) perspective views; and c) the autonomous mobile robot.

The robot navigates in an indoor environment, in this case an apartment. This environment was chosen on account of the existence of features which help the recognition and navigation of the robot. In the topological map, we presented the nodes numbered 1 to 6. We took pictures of the strategic points corresponding to classes numbered 1 to 15. The edges are the possible paths of the robot during navigation. Fig. 4 shows the detailed diagram of the real environment.

During navigation, the robot uses the vision-based localization system to locate itself. The system begins to work with the capture of images. An omnidirectional image is obtained through the Dot, a lens that can shoot 360 degree pictures. Once obtained the digital image, is analyzed by the system, which begins with feature extraction. In this paper, Central Moments, GLCM, Hu moments, LBP and Statistics Moments are evaluated. The feature vectors are presented to a classifier to perform the recognition of the environment. In this step, six machine learning techniques are used in the experiments. Both SIFT and SURF were evaluated separately. Fig. 3 shows the information flowchart.

flow_info

Fig. 3 . Proposed system flowchart. (1) The omnidirectional image is captured and the feature vector is created through feature extraction techniques. (2) The feature vectors are presented to a classifier to perform the recognition of the environment. (3) Cognition. (4) Motion control.

Following the image database:

  • Virtual image database – click here for download
  • Real image database – click here for download

This approach is being published on Expert System with Application. After that, these images will be released for use and citation of the article will be available below in order to be properly referenced when using this database.