You are here

Pattern and Landmark Detection

  • Detection of logos with different scales
  • Lift sign recognition
  • Lift botton detection and finger tip tracking

Related Projects

Related Activities

This activity focuses on the development of algorithms for the detection of landmarks naturally embedded in the scene.

Our Visual Target Detector aims at detecting and localizing in the 3d world a specified target within the visual range of a Smartphone camera. A target can be single- or multi-parts and its structure is encoded as a custom description file (xml format). The description file stores information about the number of sub-parts, their shape/size in the real world (represented by polylines) and the spatial relationships between them. The different parts are assumed to be co-planar. For each part, a specialized detection routine, if developed, can be specified in the xml file.

The Visual Target Detector works as follows:

  • All the available specialized routines are applied to the input camera image to obtain a list of candidates for each single part (e.g. a yellow square with a black icon inside);
  • Combinations of candidates are analysed in order to select the one most compatible with the target structure, taking into account the homography that maps the coordinates of the candidate regions to the real world one;
  • The homography corresponding to the best combination is used to localize each part of the input image and to estimate the pose of the camera with respect to the target (using known camera calibration data).

The introduction of a multi-part target enables the detector to be paricularly robust to partial occlusions, a useful feature especially if the target is not fully framed or in the case of interaction user-target (lift buttons can be covered by the user’s hand).

Using fast template matching methods, text in scene algorithms and skin detection techniques, we have developed specialized routines devoted to the detection of several targets, some of them composed of a single part and other composed of several parts.

Publications

Chippendale P., Tomaselli V., D'Alto V., Urlini G., Modena C.M., Messelodi S., Strano M., Alce G., Hermodsson K., Razafimahazo M., Michel T., Farinella G., Personal Shopping Assistance and Navigator System for Visually Impaired People, ECCV 2014 Workshops, Zurich, Switzerland, LNCS 8927/2014, pp. 375-390