In one implementation, a method includes generating a set of candidate objects based at least in part on a set of image data, where the set of image data includes image data from a plurality of image sources. The method also includes generating a set of visual signatures, wherein each of the visual signatures in the set of visual signatures characterizes a candidate object in the set of candidate objects. The method further includes transforming at least two candidate objects in the set of candidate objects into a single object according to a determination that correspondence between visual signatures for the at least two candidate objects satisfies one or more correspondence criteria.