Figure 1 - uploaded by Christoffer Brax
Content may be subject to copyright.
Source publication
In this paper we propose an approach for detecting anomalies in data from visual surveillance sensors. The approach includes creating a structure for representing data, building ldquonormal modelsrdquo by filling the structure with data for the situation at hand, and finally detecting deviations in the data. The approach allows detections based on...
Context in source publication
Citations
... To detect anomalies, machine learning or deep learning models can be trained using only the data from normal observations (that are generally abundantly available) and these algorithms can then flag any significant deviations as anomalous behaviour [14]. Computer vision techniques have been successfully used in identifying anomalous behaviours in homes [15], crowded scenes [16] and public areas [17]. There has also been a lot of work in the general field of video based anomaly detection using deep learning methods [18], [19]. ...
Behavioural symptoms of dementia present a significant risk within Long Term Care (LTC) homes, which face difficulties supporting residents and monitoring their safety with limited staffing resources. Many LTC facilities have installed video surveillance systems in common areas that can help staff to observe residents; however, typically these video streams are not monitored. In this paper, we present the development of a computer vision algorithm to use these video streams to detect episodes of clinically important agitation in people with dementia. Given that episodes of agitation are rare in comparison to normal behaviours, we formulated this as an anomaly detection problem. This involves using the video camera to monitor the scene rather than tracking individuals. We developed a customized spatio-temporal convolution autoencoder that is trained on the normal behaviours and then identified agitation during testing as anomalous behaviour. We present a proof-of-concept using video data collected from a specialized dementia unit and annotated for agitation events. We trained the unsupervised neural network on approximately 24 hours of normal activities and tested on 11 hours of videos containing both normal activities and agitation events, and obtained an area under the curve of the receiver operating characteristic curve of 0.754. This research paves the way for leveraging existing surveillance infrastructure in LTC and other mental health settings to detect agitation or aggression, with the potential for improved health and safety.
... To detect anomalies, machine learning models can be trained using only the data from normal observations (that are generally abundantly available) and flagging any significant deviations as anomalous behaviour [13]. Computer vision techniques have been successfully used in identifying anomalous behaviours in homes [14], crowded scenes [15] and public areas [16]. There has also been a lot of work in the general field of video based anomaly detection using deep learning methods [17], [18]. ...
Behavioural symptoms of dementia present a significant risk within Long Term Care (LTC) homes, which face difficulties supporting residents and monitoring their safety with limited staffing resources. Many LTC facilities have installed video surveillance systems in common areas that are meant to help staff observe residents; however, typically these video streams are not monitored. In this paper, we present the first use of these video streams to detect episodes of clinically important agitation in people with dementia. Given that episodes of agitation are rare in comparison to normal behaviours, we formulated this as an anomaly detection problem. This involves using the video camera to monitor the scene rather than tracking individuals. We developed a customized spatio-temporal convolution autoencoder that is trained on the normal behaviours and then identified agitation during testing as anomalous behaviour. We present proof-of-concept using video data collected from a specialized dementia unit and annotated for agitation events.
We trained the unsupervised neural network on approximately 24 hours of normal activities and tested on 11 hours of videos containing normal activities and agitation events, and obtained an area under the curve of receiver operating characteristic curve of 0.754. This research paves the way for leveraging existing surveillance infrastructure in LTC and other mental health settings to detect agitation or aggression, with the potential for improved health and safety.
... Humans can identify normal and abnormal events by manually inspecting short video frames. However, it is expensive to manually identify abnormal behavior from a large amount of data generated every day [2]. It is challenging to give meticulous attention over a long period. ...
Identifying anomalous motion behavior in video sequences is a challenging task. Manual annotation of a large number of surveillance videos is time-consuming because of the limited human brain's visual attention. This work presents a new framework to detect abnormalities from unlabeled videos using motion patterns for the normal and abnormal event. This paper proposed an unsupervised hierarchical agglomerative clustering technique for finding the abnormal behavior motion patterns. Dense trajectories of feature points were extracted and grouped into feature points for different interval groups with characteristics of the feature points' motion speed. With results from partitioning interval groups by hierarchical clustering, anomalous motion patterns were localized in surveillance video sequences. We performed experiments on publicly available datasets containing different abnormal samples. The experimental results showed that the proposed framework achieved the highest frame-level accuracy of 96.68% for the UMN dataset. The experiment has achieved the highest rate of detection (up to 98.63%) for UCSD pedestrian datasets. The proposed framework has achieved outstanding performance in both pixel level and frame level evaluation.
... Humans are able to learn, interpret and predict complex interactions and distinguish scenes. The huge amount of data are being generated from existing surveillance systems which makes difficult for human to classify complex events or manually label a video [1]. ...
Surveillance cameras are widely being used in public places for security and monitoring purposes. Detecting
abnormal motion pattern from surveillance video sequences is challenging task. Most of the existing methods
are based on supervised technique. Supervised method groups feature points into normal and abnormal
motion pattern using classifier. But anomalous event are contextual so this paper focuses on unsupervised
learning method of finding abnormal motion pattern. Contextual abnormality can not be detected by supervised
method in every surveillance video sequences. This proposed approach works without the need of training
phase. By extracting trajectory features by dense optical flow, speed of moving objects are taken into
consideration for unsupervised motion pattern in video sequences. K-means clustering approach is simple to
implement and computationally efficient. By applying such clustering method, dominant motion group and
anomalous group are well separated. Experimental results demonstrate this proposed approach outperforms
the state-of-art approaches on standard dataset.
... Abnormality or anomaly can be realized by using normal activities and depends on the approaches used to classify them. Three broad categories such as supervised, semi-supervised and un-supervised are used to construct the model [12][13][14]. ...
Detecting abnormal activities is a crucial research topic nowadays because of its wide variety of applications in such as security monitoring, video surveillance and healthcare applications. The proposed method is used to distinguish between normal and abnormal human activities. Two-dimensional visual saliency map is created from color video sequences and used for further processing. Selective spatio-temporal interest point (STIP) detector is used to extract interest point features from saliency. 3D Image gradients are calculated using intensity patches to describe STIPs and feature vector is computed by quantizing them. The activities are described finally using bag-of-features representation. Support Vector Machine is used as a classifier to distinguish between normal and abnormal activities. The performance of the system is evaluated using UR fall Dataset and dataset S provided by Le2i CNRS that shows significant accuracy.
... In the last decades, various applications in the computer vision field have emerged like face recognition [1][2][3] , events detecting [4,5] , human tracking [6][7][8][9] driving assistance [10,11] , robots control [12,13] and Video surveillance. This latter is an important task for many crucial activities like strengthening public safety or protecting important infrastructures. ...
Most object tracking methods applied in the video surveillance field are based on the prior pattern recognition of the moving objects. These methods are not adequate for tracking many different objects at the same time because the pattern of every moving object should be predefined. Thus, this paper introduces a new method to overcome this problem. Indeed, a new real time approach is established based on the particle filter and background subtraction. This approach is able to detect and track automatically, multiple moving objects without any learning phase or prior knowledge about the size, the nature or the initial position. An experimental study is performed over several video test sets. The obtained results show that the new method can successfully handle many complex situations. A comparison with other methods reports that the proposed approach is more advantageous in detecting objects as well as tracking them.
... Event detection in the field of automatic video surveillance has gained a growing interest [1]. The huge amount of data generated by existing surveillance systems in public areas requires the development of intelligent solutions that can avoid information overload for the users [2]. In particular, in the context of a crowd image analysis problem, it is desirable to develop on-line algorithms that reliably detect abnormal events in real-time. ...
Automatically detecting events in crowded scenes is a challenging task in Computer Vision. A number of offline approaches have been proposed for solving the problem of crowd behavior detection, however the offline assumption limits their application in real-world video surveillance systems. In this paper, we propose an online and real-time method for detecting events in crowded video sequences. The proposed approach is based on the combination of visual feature extraction and image segmentation and it works without the need of a training phase. A quantitative experimental evaluation has been carried out on multiple publicly available video sequences, containing data from various crowd scenarios and different types of events, to demonstrate the effectiveness of the approach.
... Specifically, anomaly detection is defined as discovering events with a low probability of occurrence from surveillance videos in computer vision community. Several approaches have been proposed in recent years and generally they can be summarized in the following three categories based on how their models are constructed, that is, supervised approaches [4][5][6][7][8][9][10], semisupervised approaches [11], and unsupervised approaches [12][13][14][15][16][17][18][19][20][21]. In real-world scenario, anomalies are usually quite rare, as its definition. ...
Automatically discovering anomalous events and objects from surveillance videos plays an important role in real-world application and has attracted considerable attention in computer vision community. However it is still a challenging issue. In this paper, a novel approach for automatic anomaly detection is proposed. Our approach is highly efficient; thus it can perform real-time detection. Furthermore, it can also handle multiscale detection and can cope with spatial and temporal anomalies. Specifically, local features capturing both appearance and motion characteristics of videos are extracted from spatiotemporal video volume (STV). To bridge the large semantic gap between low-level visual feature and high-level event, we use the middle-level visual attributes as the intermediary. And these three-level framework is modeled as an extreme learning machine (ELM). We propose to use the spatiotemporal pyramid (STP) to capture the spatial and temporal continuity of an anomalous even, enabling our approach to cope with multiscale and complicated events. Furthermore, we propose a method to efficiently update the ELM; thus our approach is self-adaptive to background change which often occurs in real-world application. Experiments on several datasets are carried out and the superior performance of our approach compared to the state-of-the-art approaches verifies its effectiveness.
... Video anomaly-detection can be realized by using a normal behavior model established from normal video data to detect unusual behaviors. Depending on the approaches applied for constructing the model, there are three broad categories, which are supervised [6][7][8][9], semi-supervised [10,11], and unsupervised [12][13][14][15]. Supervised-based methods learn the normal and abnormal behavior models via labeled video data, and can detect anomalies defined beforehand in the training phase. ...
... It belongs to a category of anomaly detection in machine learning, and such problem shares a common assumption that the de¯nition of anomaly is not explicitly given, and the objects occurring occasionally or rarely are supposed to be anomalous. The approaches for anomaly detection in existing literature can be classi¯ed into three categories: supervised, 3,9,26,30 semi-supervised 16,43,53 and unsupervised. 18,19,49,51 Supervised-based methods learn the normal and abnormal behaviors models from labeled training data, then detect anomalies via learned models. ...
In this paper, we propose a new approach for anomaly detection in video surveillance. This approach is based on a nonparametric Bayesian regression model built upon Gaussian process priors. It establishes a set of basic vectors describing motion patterns from low-level features via online clustering, and then constructs a Gaussian process regression model to approximate the distribution of motion patterns in kernel space. We analyze different anomaly measure criterions derived from Gaussian process regression model and compare their performances. To reduce false detections caused by crowd occlusion, we utilize supplement information from previous frames to assist in anomaly detection for current frame. In addition, we address the problem of hyperparameter tuning and discuss the method of efficient calculation to reduce computation overhead. The approach is verified on published anomaly detection datasets and compared with other existing methods. The experiment results demonstrate that it can detect various anomalies efficiently and accurately.
