Will the Pedestrian Cross? A Study on Pedestrian Path Prediction

Abstract

Abstract-Future vehicle systems for active pedestrian safety will not only require a high recognition performance but also an accurate analysis of the developing traffic situation. In this paper, we present a study on pedestrian path prediction and action classification at short subsecond time intervals. We consider four representative approaches: two novel approaches (based on Gaussian process dynamical models and probabilistic hierarchical trajectory matching) that use augmented features derived from dense optical flow and two approaches as baseline that use positional information only (a Kalman filter and its extension to interacting multiple models). In experiments using stereo vision data obtained from a vehicle, we investigate the accuracy of path prediction and action classification at various time horizons, the effect of various errors (image localization, vehicle egomotion estimation), and the benefit of the proposed approaches. The scenario of interest is that of a crossing pedestrian, who might stop or continue walking at the road curbside. Results indicate similar performance of the four approaches on walking motion, with near-linear dynamics. During stopping, however, the two newly proposed approaches, with nonlinear and/or higher order models and augmented motion features, achieve a more accurate position prediction of 10-50 cm at a time horizon of 0-0.77 s around the stopping event.

Using road topology to improve cyclist path prediction

“…For pedestrians, various approaches focus first on classifying current traffic behavior [2], [4], [13], which can inform future behavioral events [6]. But predictive models of a pedestrian's path must represent spatial uncertainty too.…”

“…Most literature on VRU path prediction focuses on pedestrians (e.g. [3], [4], [5], [6]), where various cues have been proposed to improve trajectory prediction, such as pedestrian attention, spatial layout, etc. Predicting cyclist tracks from a moving vehicle is also challenging, as cyclists move fast, and can be observed for long durations where high-level behavior results in distinct paths, e.g.…”

Context-Based Pedestrian Path Prediction

“…ADF can be applied to discrete state DBNs, known as Boyen-Koller inference [8], and more generally to mixed discrete-continuous state spaces with conditional Gaussian posterior [22]. Interacting Multiple Model KF [7] is related to ADF for SLDS, as it mixes the states of several KF filters running in parallel, and has been applied for path prediction in the intelligent vehicle domain [18,30].…”

“…Within the class of non-parametric methods for path prediction and action classification, [18] recently proposed two non-linear, higher order Markov models to estimate whether a crossing pedestrian will stop at the curbside, one using Gaussian Process Dynamical Models (GPDM), and one using Probabilistic Hierarchical Trajectory Matching (PHTM). Both models use dense optical flow features in the pedestrian bounding box, in addition to the positional information.…”

“Pedestrian in the loop”: An approach using virtual reality

“…There exists often no pedestrian perception stimuli unit. Exceptions are long-term studies, where pedestrians appear randomly and the type of experiment is observation inspired [17].…”