In a passive radio frequency identification (RFID) System, tags collision is caused by multiple tags communicating with the reader simultaneously, which dramatically influences the system efficiency. Therefore, researching on anti-collision algorithms to reduce the collisions and increasing the system efficiency becomes a hotspot. In this paper, we focus on the passive RFID tags and first propose a dynamical frame slotted ALOHA-based estimation method to predict the unknown population of tags. Based on this, an adjustment strategy is proposed to dynamically modify the frame size as the population of tags changes. In this adjustment strategy, a segment fashion is proposed to instead the slot-by-slot fashion; the former one could efficiently increase the reading speed and ensure the system throughput. As a result, the proposed algorithm could reach to the throughput of 35% in average with a stable estimation accuracy of 80%. The simulation shows that our algorithm also outperforms the other algorithms in the identification speed (tags/s).