As a promising technology for monitoring and tracing the vehicle flows and human activities, radio frequency identification (RFID) has received much attention in database community. k-nearest neighbor (k-NN) query over RFID monitored objects is one of the most important spatio-temporal queries used to support valuable information analysis. Different from the constraint-free space and constraint-based space, however, RFID monitoring scenario is usually merged into a semi-constraint space, which desires new data storage and distance evaluation strategies. Furthermore, the uncertainty of the monitored object locations challenges the query semantics and processing methods. In this paper, the concept of semi-constraint space is proposed, and the RFID-based semi-constraint space model is analyzed. Based on the semi-constraint space, three models and algorithms are proposed to estimate the probable k-NN results given a dynamic query point. Some special indexing techniques are adopted to speed the query. The experiment evaluates the efficiency and accuracy of the proposed algorithms and proves the effectiveness of relative methods.