Augmented vision and visual search

The video presents the visual recognition system based on a database of Geo-positioned photos of the itinerary. When the camera of the car frames a picture in the viewfinder, the system search it in the GPS photo database.

The actual effectiveness of the visual search solutions is not good enough for car navigation: it’s too slow and too inaccurate.

Our challenge is to improve the effectiveness of the visual search restricting the search field to pictures useful to our goal.The problem is to identify a database of target picture similar to the frames shooted by the camera on the car. An important source of these picture is street view, because its photocameras are collocated over the car, and the frames are very similar to the one taken from our videocamera collocated on the car. But the selection of the pictures can be made also using location based social network.

The Visual search need many target pictures to recognize an object in reality. For example, to recognize fiera Milano building, collocated in the itinerary shown by our demonstrator, we nedd a rich database of target pictures of Fiera Milano.