Hunting Nessie - Real-time abnormality detection from webcams

Abstract

We present a data-driven, unsupervised method for unusual scene detection from static webcams. Such time-lapse data is usually captured with very low or varying framerate. This precludes the use of tools typically used in surveillance (e.g., object tracking). Hence, our algorithm is based on simple image features. We define usual scenes based on the concept of meaningful nearest neighbours instead of building explicit models. To effectively compare the observations, our algorithm adapts the data representation. Furthermore, we use incremental learning techniques to adapt to changes in the data-stream. Experiments on several months of webcam data show that our approach detects plausible unusual scenes, which have not been observed in the data-stream before.

Temporal Relations in Videos for Unsupervised Activity Analysis

“…8 we show eight illustrative abnormal events that are detected among more than 250, 000 evaluated frames. We detect similar anomalies as reported in [3], such as the ones in the first line of Fig. 8.…”

Exploratory analysis of time-lapse imagery with fast subset PCA

“…However, these papers leverage similarity scores across the entire data set, and so cannot perform anomaly detection on either spatial or temporal subsets of the original data set. In this paper, we develop a system for selecting exemplar images which give similar results to the abnormalities presented in [2], with the added ability to quickly analyze arbitrary spatiotemporal subsets of the original data.…”

“…Other papers [2,18] describe anomaly detection frameworks on long-term webcam images. These systems attempt to score the "normalcy" of the current frame of a webcam image in terms of previously-seen images, and can reliably select the few most anomalous images from a longterm time-lapse sequence.…”

Improved BING Method and Its Application in Object Detection

“…Some of them are based on frequency [4,10,12,13], like Siva's unsupervised Frequency -based Saliency Computing Method [4], which uses the negative correlation between saliency and image patch frequency to compute the image saliency map. The advantage of this method is that the saliency map is accurate and unsupervised.…”