Learning Temporal Regularity in Video Sequences

TLDR

Recognizing meaningful activities in long videos is difficult because of ambiguous definitions and scene clutter. The study aims to learn a generative model of regular motion patterns in videos with minimal supervision. They develop two autoencoder approaches: one using handcrafted spatio‑temporal features with a fully connected autoencoder, and another end‑to‑end fully convolutional autoencoder that jointly learns features and classifiers. The models successfully capture regularities across datasets and achieve competitive anomaly‑detection performance in both qualitative and quantitative evaluations.

Abstract

Perceiving meaningful activities in a long video sequence is a challenging problem due to ambiguous definition of 'meaningfulness' as well as clutters in the scene. We approach this problem by learning a generative model for regular motion patterns (termed as regularity) using multiple sources with very limited supervision. Specifically, we propose two methods that are built upon the autoencoders for their ability to work with little to no supervision. We first leverage the conventional handcrafted spatio-temporal local features and learn a fully connected autoencoder on them. Second, we build a fully convolutional feed-forward autoencoder to learn both the local features and the classifiers as an end-to-end learning framework. Our model can capture the regularities from multiple datasets. We evaluate our methods in both qualitative and quantitative ways - showing the learned regularity of videos in various aspects and demonstrating competitive performance on anomaly detection datasets as an application.