Objectives  Visible landmarks and invisible landmarks are important aids for research into and design of applications for some target route.

  Methods  We propose a pedestrian relative positioning method by fusing the visible landmarks and invisible landmarks, considering the data variance in different environ‐ ments and the pedestrian customary behavior during pedestrian positioning. Firstly, the invisible landmarks (e.g., magnetic changes, WiFi(wireless fidelity) updates) along the target route are detected by smartphone sensors, and the evidence framework is built by segmenting the target route with data characteristics of sen‐ sors. Then the salient visible landmarks could be detected, and the relative spatial relations between land‐ marks and panoramas could be derived based on their coordinates.Secondly, the probability values of pedes‐ trians in the road segments are respectively obtained, based on the similarly of the real‐time sensor data and sensor data from each of the evidence framework. And the relative azimuth relations of landmarks in the panoramas could be updated instantly. Finally, based on the Bayesian probability fusion method, the pedes‐ trian positioning results could be computed through fusing the results of sensors and panoramas.In detail, the probability values of pedestrians in the road segments will be recalculated based on the panoramic image results.

  Results  The experimental results demonstrate that the proposed method could improve the posi‐ tioning accuracy in a single pedestrian walking environment by fusing multi‐source data.In an environment with fewer sensor features, the accuracy achieved by this method increases by 12.78%, which is higher than that of the invisible landmark‐based method.

  Conclusions  The pedestrian relative positioning method not only solves the problems of sensor instability and less sensor features, but also improves the positioning accuracy.