Syntactic matching of pedestrian trajectories for behavioral analysis

Abstract

In the present work we propose a new approach to dynamically characterize trajectories for a syntactic spatiotemporal alignment that can be applied in the context of behavioral analysis and anomalous activity detection. The developed architecture is based on a symbolic representation of the trajectory, exploiting the framework of the so-called editdistance. The acquired trajectory samples are filtered to identify the most significant spatio-temporal discontinuities: these key points are converted into a string-based domain where the matching of trajectory pairs can be expressed in terms of global alignment between symbols, similarly to DNA string matching algorithms. The extraction, characterization and alignment of trajectories have been tested in different environments, demonstrating the reliability of the achieved results and the viability of the solution for video surveillance and domotics applications.

Hierarchical Matching of 3D Pedestrian Trajectories for Surveillance Applications

“…The segmentation strategy we have adopted in this work is an evolution of the algorithm proposed in [1], and the main points of innovation can be summarized in (i) the substantial improvements of the features quantization to better represent pedestrian trajectories and (ii) the introduction of the height y to achieve a full 3D representation.…”

“…At each trajectory sample P i j = (x i j , y i j , z i j , t j ) the procedures described in [1] are evaluated to determine stops and direction variations according to temporal and spatial constraints. At the end of the segmentation process, the In this representation each segment S i m is characterized by its orientation θ i m , the height of the centroid y i m , its velocity v i m , and the temporal interval Δt m .…”

“…With respects to the speed variation range reported in Table I, one level has been allocated to the null velocity (stop), while the upper level covers the range from v 3 up to v max , which specifies the maximum possible velocity. A similar approach is applied to time, where the maximum level is set to the threshold used to detect stops (see [1] for details). The quantization of the 3D position of the object is uniformly discretized along y at regular intervals of 30cm, while the direction in x − z has been quantized considering a new level every 30 • .…”

“…The scope of the work is to derive anomalies in common actions through a 3D analysis enriched by the information retrieved from additional features like speed and time. A preliminary version about trajectory analysis in the 2D domain has been presented in [1]. Here we enhance the system by introducing the third dimension to improve the trajectory classification in a hierarchical fashion, making it possible to deploy a tool for automatic learning of new movement patterns.…”

Syntactic Matching of Trajectories for Ambient Intelligence Applications

“…Starting from a preliminary study proposed in [9], we present in this work a complete representation and matching framework, and provide an in-depth description of the relevant processing techniques and a thorough experimental validation, also in comparison with state-of-art approaches of the same class.…”