Visual Heading-Aided Pedestrian Navigation Method Based on Factor Graph in Indoor Environment

The main error in traditional inertial pedestrian navigation is caused by unobservable heading errors with only zero velocity constraints. In this article, a human posture model-aided optimized fusion framework based on dominant events for pedestrian navigation is proposed to precisely locate pedestrians. Optimization is performed only at the moment of dominant events to reduce computational effort and improve the robustness. This model can fit the preferred heading information between two dominant events for fusion to improve accuracy. The optimization-based framework can also reduce the nonlinear error by relinearizing the state and associating constraints to the historical state to estimate the biases better. Besides, a hybrid visual heading angle estimation method is proposed by leveraging the Manhattan world hypothesis to obtain a drift-free heading angle with a monocular camera. The hypothesis is searched in the designated region based on histogram quality assessment that can improve the robustness and significantly reduce the computational amount. With low-cost devices, the multiple indoor experiments show that the proposed method can achieve 0.11% positioning accuracy comparable to traditional algorithms, which require numerous conditions to be met in good environments, and achieving over 70% improvement in accuracy over traditional methods in complex environments.