In order to solve data collision problem caused by multiple tags sending information to receptor in underground personnel positioning system, an improved algorithm of binary exponential backoff was proposed. The algorithm sets two threshold values by way of multiple increase and linear decrease to adjust collision window, and makes different update rules for value of backoff generator according to different network traffic. At the same time, it synchronizely optimizes the window value to make the tag adaptively and quickly access channel. The tests showed that the maximum number of concurrent identification number of the improved algorithm is 150, and the maximum displacement speed is 10 m/s, both are superior to the classical binary exponential backoff algorithm. The algorithm improves data transmission rate and reduces card leakage rate, which is an effective solution for anti-collision problem of underground multi-target recognition.