Advances in Mobile Mapping Technology

Publication from Digital

Paletta L., Fritz G., Amlacher K., Luley P., Stelzl H., Almer A.

Taylor & Francis, Leiden, Netherlands , 2008


In the context of mapping and mobile vision services, the recognition of objects of the urban infrastructure plays an important role. E.g., the recognition of buildings can foster pointers to relevant information spaces, such as, annotation services, it can provide a semantic index for scene understanding, and can enable more efficient navigation by direct reference to landmark buildings. The presented work provides a generic technology for the recognition of urban objects, i.e., buildings, in terms of a reliable mobile vision service. The presented detection system grounds recognition on a Maximum A Posteriori based decision making on weak object hypotheses from local descriptor responses in the mobile imagery. We present an improvement over standard image descriptors by selecting only informative keys for descriptor matching. Selection is applied first to reduce the complexity of the object model and second to accelerate detection by selective attention. We trained a decision tree to rapidly and efficiently estimate an image descriptor’s posterior entropy value, retaining only those keys for thorough analysis and voting with high information content. The experiments were performed on typical, low quality mobile phone imagery on urban tourist sites under varying environment conditions (changes in scale, viewpoint, illumination, varying degrees of partial occlusion). We demonstrate in this challenging outdoor object detection task the superiority in using informative keys to standard descriptors using a publicly available mobile phone image database, reporting increased reliability in object/background separation, accurate object identification, and providing a confidence quality measure that enables a highly stable mobile vision service. We show further results on imagery captured from a mobile mapping van that demonstrate the capability to localize complex objects of interest in un-calibrated imagery within urban environments.

Keywords: mobile vision systems, object recognition, building detection, urban environments