Anomaly detection in video sequences: A benchmark and computational model

Abstract

Anomaly detection has attracted considerable search attention. However, existing anomaly detection databases encounter two major problems. Firstly, they are limited in scale. Secondly, training sets contain only video-level labels indicating the existence of an abnormal event during the full video while lacking annotations of precise time durations. To tackle these problems, we contribute a new Large-scale Anomaly Detection (LAD) database as the benchmark for anomaly detection in video sequences, which is featured in two aspects. 1) It contains 2000 video sequences including normal and abnormal video clips with 14 anomaly categories including crash, fire, violence etc. with large scene varieties, making it the largest anomaly analysis database to date. 2) It provides the annotation data, including video-level labels (abnormal/normal video, anomaly type) and frame-level labels (abnormal/normal video frame) to facilitate anomaly detection. Leveraging the above benefits from the LAD database, we further formulate anomaly detection as a fully supervised learning problem and propose a multi-task deep neural network to solve it. We firstly obtain the local spatiotemporal contextual feature by using an Inflated 3D convolutional (I3D) network. Then we construct a recurrent convolutional neural network fed the local spatiotemporal contextual feature to extract the spatiotemporal contextual feature. With the global spatiotemporal contextual feature, the anomaly type and score can be computed simultaneously by a multi-task neural network. Experimental results show that the proposed method outperforms the state-of-the-art anomaly detection methods on our database and other public databases of anomaly detection.

Self-supervised Learning for Unintentional Action Prediction

“…To this end, we combine a global temporal contrastive loss with a pair-wise video clip order prediction task. We show that our proposed approach outperforms all previously proposed self-supervised models for unintentional action prediction and achieves state-of-the-art results on two datasets of unintentional actions: the Oops dataset [9] and LAD2000 [32].…”

“…We conducted our experiments on two datasets with unintentional or anomalous videos: Oops [9] and LAD2000 [32]. We present the results on the LAD2000 dataset in the supplementary material.…”

Anomaly detection in surveillance videos based on H265 and deep learning

“…Wan et al [31] Wu et al [34] proposed a dual branch network using multi-detail concepts in both the temporal and spatial dimensions as input. C3D was used to extract the features, and namely spatio-temporal (ST-UCFcrime) was used to implement them on a new dataset (ST-UCF-crime) that annotates spatial-temporal bounding boxes for unusual occurrences in UCF crime.…”

Imitating Emergencies: Generating Thermal Surveillance Fall Data Using Low-Cost Human-like Dolls

“…Outdoor fall detection needs to be feasible in non-static environments with changing backgrounds, light conditions, density and clutter. This necessitates more robust approaches combining video motion detection and temporal features using three dimensional LSTM and ConvLSTM [ 12 , 40 ], extracting depth from single images using Markov Random Fields and human detection using particle swarm optimization [ 41 ] or body posture analysis using two-branch multi-stage CNNs [ 42 ].…”