Classifying spatiotemporal object trajectories using unsupervised learning in the coefficient feature space

Abstract

This paper proposes a novel technique for clustering and classification of object trajectory-based video motion clips using spatiotemporal function approximations. Assuming the clusters of trajectory points are distributed normally in the coefficient feature space, we propose a Mahalanobis classifier for the detection of anomalous trajectories. Motion trajectories are considered as time series and modelled using orthogonal basis function representations. We have compared three different function approximations -least squares polynomials, Chebyshev polynomials and Fourier series obtained by Discrete Fourier Transform (DFT). Trajectory clustering is then carried out in the chosen coefficient feature space to discover patterns of similar object motions. The coefficients of the basis functions are used as input feature vectors to a SelfOrganising Map which can learn similarities between object trajectories in an unsupervised manner. Encoding trajectories in this way leads to efficiency gains over existing approaches that use discrete point-based flow vectors to represent the whole trajectory. Our proposed techniques are validated on three different datasetsAustralian sign language, hand-labelled object trajectories from video surveillance footage and real-time tracking data obtained in the laboratory. Applications to event detection and motion data mining for multimedia video surveillance systems are envisaged.

Goal-based trajectory analysis for unusual behaviour detection in intelligent surveillance

“…The second question is addressed through experiments with Piciarelli et al's synthetic dataset [42] and Naftel and Khalid's real-world laboratory dataset [39]. Hence, all three datasets used in these experiments are published by third-party researchers working on trajectory-based unusual behaviour detection.…”

“…The following chapter elaborates on the contributions of this thesis to the goal-based approach. Johnson and Hogg [25] Owens and Hunter [40] Stauffer and Grimson [48] Hu et al [19] Makris and Ellis [32,33] Fernyhough et al [14] Piciarelli and Foresti [41] Piciarelli et al [42] Junejo et al [26] Naftel and Khalid [39] Jiang et al [24] Sillito and Fisher [47] Dee and Hogg [11] Proposed Method…”

“…Naftel and Khalid represented trajectories using function approximations, including least square polynomial, Cheybyshev polynomial, and Discrete Fourier Transform [39].…”

“…Further validation of the proposed method was performed using the real-world dataset developed by Naftel and Khalid [39]. The trajectories in Naftel and Khalid's dataset are higher tendency to issue false alarms on this dataset.…”

“…The [39]. dataset consists of 262 normal trajectories from surveillance video captured over an hour and six trajectories of volunteers instructed to "behave in a suspicious fashion" though without any knowledge of Dee and Hogg's goal-based method of analysis [11].…”