Reliable Heading Tracking for Pedestrian Road Crossing Prediction Utilizing Commodity Devices > 자유게시판

Pedestrian heading monitoring enables functions in pedestrian navigation, traffic security, and accessibility. Previous works, utilizing inertial sensor fusion or machine studying, are restricted in that they assume the telephone is fastened in specific orientations, hindering their generalizability. We suggest a new heading monitoring algorithm, the Orientation-Heading Alignment (OHA), which leverages a key insight: people have a tendency to carry smartphones in certain ways because of habits, reminiscent of swinging them whereas walking. For each smartphone attitude during this motion, OHA maps the smartphone orientation to the pedestrian heading and learns such mappings effectively from coarse headings and smartphone orientations. To anchor our algorithm in a practical scenario, we apply OHA to a difficult job: predicting when pedestrians are about to cross the street to enhance road user safety. Particularly, using 755 hours of walking information collected since 2020 from 60 individuals, we develop a lightweight model that operates in real-time on commodity units to predict street crossings. Our analysis shows that OHA achieves 3.4 occasions smaller heading errors across 9 situations than existing strategies.

Furthermore, OHA allows the early and correct detection of pedestrian crossing conduct, issuing crossing alerts 0.35 seconds, on common, earlier than pedestrians enter the road range. Tracking pedestrian heading entails repeatedly tracking an individual’s dealing with route on a 2-D flat aircraft, typically the horizontal plane of the worldwide coordinate system (GCS). Zhou et al., 2014). For example, a pedestrian might be walking from south to north on a highway whereas swinging a smartphone. In this case, smartphone orientation estimation would point out the iTagPro Device’s dynamic orientation relative to the GCS, commonly represented by Euler angles (roll, pitch, yaw). Alternatively, monitoring pedestrian heading ought to precisely show that the pedestrian is shifting from south to north, regardless of how the smartphone is oriented. Existing approaches to estimating pedestrian heading via IMU (Inertial Measurement Unit) make use of a two-stage pipeline: first, they estimate the horizontal plane utilizing gravity or magnetic fields, and then integrate the gyroscope to track relative heading changes (Manos et al., 2018; Thio et al., 2021; Deng et al., 2015). These approaches hinge on a vital assumption: the cellphone should remain static relative to the pedestrian body.

We suggest a brand new heading monitoring algorithm, Orientation-Heading Alignment (OHA), which leverages a key perception: folks have a tendency to carry smartphones in certain attitudes attributable to habits, whether swinging them whereas strolling, stashing them in pockets, or putting them in bags. These attitudes or relative orientations, defined as the smartphone’s orientation relative to the human body moderately than GCS, primarily rely on the user’s habits, characteristics, and even clothes. As an illustration, no matter which route a pedestrian faces, they swing the smartphone of their habitual manner. For iTagPro Device every smartphone perspective, OHA maps the smartphone orientation to the pedestrian heading. Because the attitudes are comparatively stable for every particular person (e.g., holding a smartphone in the appropriate hand and swinging), it is feasible to study the mappings efficiently from coarse headings and smartphone orientation. Previous analysis (Liu et al., 2023; Yang et al., 2020; Lee et al., 2023) has famous the same perception but adopted a unique strategy for heading tracking: accumulating IMU and accurate heading info for multiple smartphone attitudes and coaching a machine studying mannequin to predict the heading.

However, attributable to system discrepancies and varying consumer behaviors, it's not feasible to assemble a machine studying model that generalizes to all potential smartphone attitudes. To anchor our heading estimation algorithm in a sensible state of affairs, we apply OHA to a difficult process: predicting when pedestrians are about to cross the road-an essential downside for improving road consumer security (T., pril; Zhang et al., 2021, 2020). This task, which requires accurate and timely predictions of pedestrian crossings, is further difficult by the numerous crossing patterns of pedestrians and the complexity of road layouts. Based on the OHA heading, we suggest PedHat, a lightweight, infrastructure-free system that predicts when a pedestrian is about to cross the closest street and points crossing alerts. PedHat incorporates a lightweight model that accepts OHA headings as inputs and operates in actual-time on person gadgets to predict highway crossings. We developed this mannequin using knowledge we collected since 2020 from 60 people, each contributing two months of traces, overlaying 755 hours of walking knowledge.