A comparative evaluation of anomaly detection algorithms for maritime video surveillance

Auslander, Bryan; Gupta, Kalyan Moy; Aha, David W.

doi:10.1117/12.883535

Cited by 22 publications

(17 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Next, we eliminate the clusters with fewer samples, based on the assumption that these smaller clusters contain mostly outlier samples. We note that the same assumption also sits at the basis of the method proposed in [5]. Nonetheless, we motivate the assumption through the following toy example.…”

Section: Two-stage Outlier Detectionmentioning

confidence: 79%

“…Moreover, their approach does not allow to set an abnormality threshold, and thus, it cannot be optimized for better precision or recall. More closely to our approach, Auslander et al [5] defined three possible assumptions (see Section 4.1 in [5]) for using clustering to detect anomalies. Interestingly, our approach is based on similar assumptions.…”

Section: Related Workmentioning

confidence: 99%

“…Interestingly, our approach is based on similar assumptions. However, their approach adopts only the first two assumptions defined in [5], while we satisfy the second assumption by eliminating smaller clusters (in the first stage), and the first and third assumptions by training a one-class SVM on each cluster (in the second stage).…”

Section: Related Workmentioning

confidence: 99%

See 2 more Smart Citations

Detecting Abnormal Events in Video Using Narrowed Normality Clusters

Ionescu

Smeureanu

Popescu

et al. 2019

2019 IEEE Winter Conference on Applications of Computer Vision (WACV)

View full text Add to dashboard Cite

We formulate the abnormal event detection problem as an outlier detection task and we propose a two-stage algorithm based on k-means clustering and one-class Support Vector Machines (SVM) to eliminate outliers. In the feature extraction stage, we propose to augment spatio-temporal cubes with deep appearance features extracted from the last convolutional layer of a pre-trained neural network. After extracting motion and appearance features from the training video containing only normal events, we apply k-means clustering to find clusters representing different types of normal motion and appearance features. In the first stage, we consider that clusters with fewer samples (with respect to a given threshold) contain mostly outliers, and we eliminate these clusters altogether. In the second stage, we shrink the borders of the remaining clusters by training a one-class SVM model on each cluster. To detected abnormal events in the test video, we analyze each test sample and consider its maximum normality score provided by the trained one-class SVM models, based on the intuition that a test sample can belong to only one cluster of normality. If the test sample does not fit well in any narrowed normality cluster, then it is labeled as abnormal. We compare our method with several state-of-the-art methods on three benchmark data sets. The empirical results indicate that our abnormal event detection framework can achieve better results in most cases, while processing the test video in real-time at 24 frames per second on a single CPU.

show abstract

Section: Two-stage Outlier Detectionmentioning

confidence: 79%

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Detecting Abnormal Events in Video Using Narrowed Normality Clusters

Ionescu

Smeureanu

Popescu

et al. 2019

2019 IEEE Winter Conference on Applications of Computer Vision (WACV)

View full text Add to dashboard Cite

show abstract

“…The traffic was investigated, and trajectories of detected vessels were further processed to maintain knowledge and conceive the representation of the normal course of movements. Auslander et al [17] supposed that different kinds of maritime traffic were characterized by different levels of complexity and presented two global and two local anomaly detection algorithms, whose performance varied depending on the maritime traffic type. Vespe et al [18] proposed a waypoint density algorithm to compress a large volume of AIS data into a list of adequate waypoints and ultimately delineate sea lanes and subsequently routes through a set of lines connecting waypoints.…”

Section: Literature Reviewmentioning

confidence: 99%

Extracting Shipping Route Patterns by Trajectory Clustering Model Based on Automatic Identification System Data

Pan

Yin

2018

Sustainability

View full text Add to dashboard Cite

Shipping route analysis is essential for vessel traffic management and relies on professional technical facilities for collecting and recording specific information about vessel behaviors. The recent Automatic Identification System (AIS) onboard has been made available to provide ship-related information for the research. However, the complexity and large quantity of AIS data overload traditional surveillance operations and increase the difficulty of vessel traffic analysis. An unsupervised approach is urgently desired to effectively convert the raw AIS data to regular shipping route patterns. In this paper, we proposed a trajectory clustering model based on AIS data to analyze the shipping routes. The whole model consists of four parts: Data preprocessing, structure similarity measurement, clustering, and representative trajectory extraction. Our model comprehensively considered the geospatial information and contextual features of ship trajectory. The revised density-based clustering algorithm could automatically classify different shipping routes with trajectory features without prior knowledge. The experimental evaluation showed the effectiveness of the proposed model by real AIS data from Port of Tianjin. The results contribute to the further understanding of shipping route patterns and assists maritime authorities and the officers in stable and sustainable vessel traffic management.

show abstract

“…Before further progressing on the description of the internal algorithms of the ETA, it should be remarked that several possible algorithms exist and can be compared through benchmarks [24], [25], [31]. It is outside the scope of this paper to build a novel algorithm or compare existing ones to identify the most efficient.…”

Section: The Event Trust Analysis (Eta)mentioning

confidence: 99%

Identification of critical situations via Event Processing and Event Trust Analysis

Itria

Kocsis-Magyar²,

Ceccarelli

et al. 2016

Knowl Inf Syst

View full text Add to dashboard Cite

In crisis management systems, situational awareness is usually at the basis of guiding the intervention process, and it is required to rapidly process data acquired from information sources on the field such as sensors or even humans. Given the variety and heterogeneity of sources and the amount of information that can be collected, together with the urgency of taking decisions, such information needs to be rapidly collected, filtered and aggregated in a form that can be used in subsequent machine-assisted decision support processes. At the same time, uncertainties in the input data or approximations in the processing phase may lead to an incorrect interpretation of the real situation in progress, which may generate mismanagements and severe consequences. This paper presents an event processor for crisis management systems that combines heterogeneous input sources to detect a critical situation. Complex event processing technology is applied for correlating data and creating events that describe the critical situation. Anomaly detection techniques are then used to analyze such events and detect possible anomalies, i.e. events not pertaining to the identified critical situation. The devised event processor creates trusted events that describe a critical situation merging inputs from heterogeneous and potentially untrusted sources. A prototype of the solution has been implemented and exercised within the crisis management system developed during the Secure! project. The experimental validation activities performed make use of different input sources, such as Twitter and sensors deployed on field (a doppler radar for people detection and accelerometers for vibrations detection). The objective of the experimental campaign is to show i) the adequacy of the solution to rapidly process the information and describe the critical situation, and ii) its capability in detecting anomalous events that could impair the accuracy of the description of the critical situation.

show abstract

A comparative evaluation of anomaly detection algorithms for maritime video surveillance

Cited by 22 publications

References 16 publications

Detecting Abnormal Events in Video Using Narrowed Normality Clusters

Detecting Abnormal Events in Video Using Narrowed Normality Clusters

Extracting Shipping Route Patterns by Trajectory Clustering Model Based on Automatic Identification System Data

Identification of critical situations via Event Processing and Event Trust Analysis

Contact Info

Product

Resources

About