This paper investigates an object detection prob-lem using a mobile robot equipped with a vision sensor. The main novelty of this work is an approach that combines localization of the robot relative to an object believed to be the target and confirmation of this object's identity. Since the position of the robot relative to the candidate target is never exactly known, we model this position by a probability distribution over a set of cells forming a decomposition of the workspace around the candidate target. By performing a series of moves the robot acquires several images and runs a target detector module on each image. Its goal is not only to reach a position where the target detector can confirm the target with high confidence (as this approach would be prone to false positives). It is also to reach a position where, with high probability, the target detector will confirm with high confidence that the candidate target is actually the target. This twofold goal reduces drastically the likelihood of false positives. The target confirmation problem is modeled as a Partially-Observable Markov Decision Process (POMDP), which is solved using Stochastic Dynamic Programming (SDP).