Research on Pedestrian Sensing Technology: Design and Applications


Student thesis: Doctoral Thesis

View graph of relations


Related Research Unit(s)


Awarding Institution
Award date5 Jan 2018


This thesis studies a crucial and applied technique in smart city designs, namely pedestrian sensing. Its primary function is to conduct step counting, i.e., counting the walking steps of a user per day, by using the motion sensors from smart devices, e.g., smart watch. As this senses the dominant and also the most primary behavior of ordinary people, an accurate pedestrian sensing design could benefit variant social aspects, such as e-health, fitness, insurance business, etc. In addition, an accurate pedestrian sensing can also be extended to estimate user’s fine-grained per-step walking length, which could further benefit useful application scenarios, like indoor navigation services. However, excessive market surveys and customer feedback indicate the unreliable and untrustworthy performance of existing pedestrian sensing designs, due to a series of unsolved applicability issues, which significantly limit this technique’s wider usage in practice and its potential market space. In this thesis, we strive to first identify and solve these challenging issues so as to advance the usability of pedestrian sensing, based on which we further leverage the ability of accurate
pedestrian sensing to develop useful applications.

The first part of this thesis focuses on an accurate and reliable pedestrian sensing design. Through our study, we find that existing pedestrian sensing solutions suffer two core unsolved challenges: vulnerability to interfering activities and parameter training without user’s intervention. To address the first issue, our initial observation unveils a composition difference between human’s walking activity and other interfering activities - interfering activities are mainly generated from rigid objects, e.g., user’s arm or hand, while walking is an aggregated activity from both the arm and body, whose movements are concurrent but relatively independent. We find that it is viable to disclose such a difference by exploring the correlation among a set of key sensor data points, after wearable sensor signals are projected to two perpendicular dimensions. In addition, to avoid user’s explicit labeling of sensory data to turn the user-specific parameter, we leverage the opportunity when user’s smartphone is on the body and we utilize the data consistence between smartwatch and smartphone to automatically train the parameter. Based on these observations, we develop a new solution which can reduce the error rate to 2% with extensive interfering activities during user’s walking.

With an accurate and reliable step counting, the second part of this thesis is to extend the pedestrian sensing to estimate user’s fine-grained per-step walking length, also known as stride estimation. The extended pedestrian sensing, with stride estimation, could directly lead to useful applications, such as indoor navigation systems. Moreover, it could also serve as functional modules to produce more sophisticated smart city applications that will be investigated in the next part of the thesis. For the stride estimation, even interfering activities
can be excluded, the motion sensors normally measure the additive effect from both the arm and body as the user walks, but stride is purely determined by body’s movement. To address this issue, we propose to decompose the gait cycle into subtle phases and leverage a geometrical relation from user’s arms, unveiled cross different phases, to extract only body’s motion for the stride estimation. Our evaluation demonstrates that the overall stride length estimation error is less than 7.0cm per step.

In the last part of this thesis, we leverage the pedestrian sensing as a core module to enable a novel and useful application for smart city designs, which could turn a user’s smartwatch to a virtual ruler. This design can estimate the size of objects that are too tall to reach or too far to touch, which will enable a rich set of applications which count on the size information of surrounding environments/objects.In this part of the thesis, we customize the pedestrian sensing technique by leveraging user’s relatively stable mobility feature to further improve the sensing performance. In addition, we also cope with a series of application-dependent challenges to fulfill the design. Extensive experimental studies in both indoor and outdoor environments show that the object size measure error is small, e.g., 8% on average.

In summary, this thesis studies a very crucial and applied sensing technique using smart devices, called pedestrian sensing, where we address a series of applicability issues to make this technique truthfully usable in practice, and also extend its functionality and demonstrate its strength to develop innovative applications. Our research output could enrich the methodologies in the wearable sensing field and also has great potentials to directly benefit the design and the space of smart city applications.

    Research areas

  • Wearable device, Mobile computing, Pedestrian sensing