The traditional Zero Velocity Updating algorithm can be used to correct the accumulated errors of the device effectively. However, as the threshold value of the traditional Zero Velocity Updating algorithm is fixed, it is only suitable for a single motion mode. When indoor pedestrian motion includes multiple motion modes, the positioning accuracy will be greatly reduced. In this paper, we propose a half-voting random forest algorithm, an adaptive Zero Velocity Updating method for multi-motion mode using half-voting Random Forest. Half-voting Random Forest Algorithm means that in the decision tree voting process, the decision tree stops voting when the voting number reaches half of the voting volume. We determined the optimal thresholds of Zero Velocity Updating value for standing still, walking, running, going upstairs and downstairs for the indoor pedestrian. Then we recognize pedestrian motion modes by Random Forest with half-voting and weighted decision trees. Finally, we assign the optimal threshold based on the motion modes for Zero Velocity Updating. In order to verify the feasibility and effectiveness of the method proposed in this paper, field experiments were carried out with the inertial navigation module developed by our laboratory. The experimental results show that when indoor pedestrians perform multi-mode motion, the positioning error is 0.5 m.