A strategy for object recognition with detection errors

This study proposes an optimal strategy for object recognition with detection errors. Detection probability improves as objects get closer to a target; yet, at a certain point it might be too late. The objective is to design an optimal strategy for recognition that guarantees accurate and timely intruder detection. The problem is formulated as a Partially Observable Markov Decision Process (POMDP): Signals from approaching objects (observations) are used to update the detection probability. We show that the optimal policy is of a Control Limit Threshold (CLT) type: The optimal policy is to continue inspection if and only if the detection probability is less than a CLT value.