Suspicious Activity Detection Research Articles

Fusing Dynamic Images and Depth Motion Maps for Action Recognition in Surveillance Systems

Background: Identification of human actions from video has gathered much attention in past few years. Most of the computer vision tasks such as Health Care Activity Detection, Suspicious Activity detection, Human Computer Interactions etc. are based on the principle of activity detection. Automatic labelling of activity from videos frames is known as activity detection. Objective: Motivation of this work is to use most out of the data generated from sensors and use them for recognition of classes. Recognition of actions from videos sequences is a growing field with the upcoming trends of deep neural networks. Methods: Automatic learning capability of Convolutional Neural Network (CNN) make them good choice as compared to traditional handcrafted based approaches. With the increasing demand of RGB-D sensors combination of RGB and depth data is in great demand. Results: This work comprises of the use of dynamic images generated from RGB combined with depth map for action recognition purpose. We have experimented our approach on pre trained VGG-F model using MSR Daily activity dataset and UTD MHAD Dataset. We achieve state of the art results. To support our research, we have calculated different parameters apart from accuracy such as precision, F score, recall. Conclusion: Accordingly, the investigation confirms improvement in term of accuracy, precision, F-Score and Recall. The proposed model is 4 Stream model is prone to occlusion, used in real time and also the data from the RGB-D sensor is fully utilized.

Deep-Facial Feature-Based Person Re-identification for Authentication in Surveillance Applications

Nowadays, a large network of cameras is predominantly used in public places which provide enormous video data. These data are monitored manually and may be utilized only when the need arises to ascertain the facts. Automating the system can improve the quality of surveillance and be useful for high-level surveillance tasks like person identification, suspicious activity detection or undesirable event prediction for timely alerts. In this paper, we proposed a model that can Re-identify a person from a single camera tracking environment. This system will automatically extract face features of the person and generate the Unique Id for each person when it enters for the first time in the monitored area. Its face features are stored in the database which will help to Re-identify the person whenever the same person appears again. The challenges faced by the system are occlusion, pose, light conditions, and face orientation. The proposed system highlights, effect of different deep neural networks for Person Re-identification and compares based on the accuracy, GPU usage, Speed, Number of faces detected by overcoming the challenges like illumination and occlusion. The advantage of the system is it doesn′t require the database of people in advance for recognition and it will be helpful for criminal identification for crime control and prevention.

Database Creation for Normal and Suspicious Behaviour Identification in ATM Video Surveillance

A database is proposed for testing real-time activities associated with Automated Teller Machine (ATM) application. The actual videos recorded during acquisition are used for testing purposes. The scenarios for different types of occurrences in ATM cabins that can lead to the detection of suspicious human activity are documented. The work in progress involves the automated segmentation of video monitoring and database output enhancement for real- time acquisition.

A Time-Frequency Based Suspicious Activity Detection for Anti-Money Laundering

Money laundering is the crucial mechanism utilized by criminals to inject proceeds of crime into the financial system. The primary responsibility of the detection of suspicious activity related to money laundering is with the financial institutions. Most of the current systems in these institutions are rule-based and ineffective (over 90 % false positives). The available data science-based anti-money laundering (AML) models to replace the existing rule-based systems work on customer relationship management (CRM) features and time characteristics of transaction behaviour. Due to thousands of possible account features, customer features, and their combinations, it is challenging to perform feature engineering to achieve reasonable accuracy. Aiming to improve the detection performance of suspicious transaction monitoring systems for AML systems, in this article, we introduce a novel feature set based on time-frequency analysis, that uses 2-D representations of financial transactions. Random forest is utilized as a machine learning method, and simulated annealing is adopted for hyperparameter tuning. The designed algorithm is tested on real banking data, proving the results’ efficacy in practically relevant environments. It is shown that the time-frequency characteristics are discriminatory features for suspicious and non-suspicious entities. Therefore, these features substantially improve the area under curve results (over 1%) of the existing data science-based transaction monitoring systems. Using time-frequency features alone, a false positive rate of 14.9% has been achieved, with an F-score of 59.05%. When combined with transaction and CRM features, the false positive rate is 11.85%, and the F-Score is improved to 74.06%.

SPEED: A Deep Learning Assisted Privacy-Preserved Framework for Intelligent Transportation Systems

Roadside cameras in an Intelligent Transportation System (ITS) are used for various purposes, e.g., monitoring the speed of vehicles, violations of laws, and detection of suspicious activities in parking lots, streets, and side roads. These cameras generate big multimedia data, and as a result, the ITS faces challenges like data management, redundancy, and privacy breaching in end-to-end communication. To solve these challenges, we propose a framework, called SPEED, based on a multi-level edge computing architecture and machine learning algorithms. In this framework, data captured by end-devices, e.g., smart cameras, is distributed among multiple Level-One Edge Devices (LOEDs) to deal with data management issue and minimize packet drop due to buffer overflowing on end-devices and LOEDs. The data is forwarded from LOEDs to Level-Two Edge Devices (LTEDs) in a compressed sensed format. The LTEDs use an online Least-Squares Support-Vector Machines (LS-SVMs) model to determine distribution characteristics and index values of compressed sensed data to preserve its privacy during transmission between LTEDs and High-Level Edge Devices (HLEDs). The HLEDs estimate the redundancy in forwarded data using a deep learning architecture, i.e., a Convolutional Neural Network (CNN). The CNN is used to detect the presence of moving objects in the forwarded data. If a movement is detected, the data is forwarded to cloud servers for further analysis otherwise discarded. Experimental results show that the use of a multi-level edge computing architecture helps in managing the generated data. The machine learning algorithms help in addressing issues like data redundancy and privacy-preserving in end-to-end communication.

Suspicious activity detection using deep learning in secure assisted living IoT environments

Children who are left alone in environments such as daycares and creches require constant monitoring and care to protect them from abuse. In this paper, we propose a novel deep learning-based method for predicting the occurrence of abnormal events using footage gathered from networked surveillance systems and notifying users of those events in an Internet of Things (IoT) environment. Sequences of images are converted to still frames and de-blurred using adaptive motion detection techniques. Then, abnormal activities are predicted using random forest differential evolution with kernel density (RFKD), and any abnormal activities that are detected cause signals to be sent to IoT devices via the MQTT protocol. The proposed work consists of a multi-classifier, deep neural network and kernel density functions. The multi-classifier is used for input classifications from the sequence of frames of videos. The deep neural network is used to learn and train the data and kernel density is used clustering and prediction of data. The novelty of the proposed work is in the dynamic nature of activity prediction. Most of the previous work in this research area concentrated on static activity prediction. The proposed work is able to support both static and dynamic activities of daycare environments. In our experimental trials, our novel method’s performance is shown to be superior to that of the ReHAR method.

Suspicious Activity Detection and Tracking through Unmanned Aerial Vehicle Using Deep Learning Techniques

Suspicious Activity Detection and Tracking through Unmanned Aerial Vehicle Using Deep Learning Techniques

Smart Security System for Suspicious Activity Detection in Volatile Areas

The latest progress in the technology has led to automation and digitization in almost every fields, and has influenced a wide scope of application. This has caused enormous amount of data flow from each sectors, where the information contained in the data acts as the important component for the progress of the single person, organization, state, country and so on. These data with valuable information can be used in the constructive and the destructive perceptive based on the hands that handle it. So protective measures become very essential for preserving the data from unwanted access. This paves for developing a system to identify the suspicious movement in the volatile areas like military regimes, hospitals and financial organizations to safe the data. The method put forward in the paper incorporates the motion sensors and the face identification system to detect the suspicious activities and report to the lawful person. The algorithm for the system was developed using the python and tested for various sets of exemplary real time video recordings to know the accuracy in the detection.

Intelligent Analysis of Arabic Tweets for Detection of Suspicious Messages

With the widespread use of messaging via social networks such as Twitter, Instagram, and Facebook, it is becoming imperative for researchers to devise intelligent systems for data analytics in the range of domains like business, health, communication, security, etc. The complex morphological and syntactic structure of Arabic sentences makes them difficult to analyze. This paper presents an intelligent system to analyze Arabic tweets for detecting suspicious messages. We acquired Arabic tweet data from micro-blogging social network Twitter via Twitter Streaming Application Programming Interface and save it in a required file format. The system tokenizes and preprocesses the tweet dataset. Manual labeling is performed on tweet dataset for suspicious (label 1) and not-suspicious (label 0) classes. The labeled tweet dataset is used to train a classifier using supervised machine learning algorithms for the detection of suspicious activities. During the testing phase, the system processes unlabeled tweet data and detects either it belongs to a suspicious or not-suspicious class. We tested the system using six supervised machine learning algorithms: (1) decision tree, (2) k-nearest neighbors, (3) linear discriminant algorithm, (4) support vector machine, (5) artificial neural networks, and (6) long short-term memory networks. A comparative analysis in terms of accuracy, execution time, and confusion matrices of the six classifiers is presented. The execution speed of ANN is lowest. In terms of predicting correct results, the SVM performs best among all the classifiers and yields 86.72% mean accuracy. The major outcomes of this work are development of labeled dataset of Arabic tweets, an intelligent behavior analysis of tweets using six machine learning algorithms to detect suspicious messages, a comparative analysis of six machine learning algorithms, and a development of a statistical benchmark that can be used for future studies about the detection of crimes on social media.

Improved Accuracy of Suspicious Activity Detection in Surveillance Video

Suspicious activity detection from surveillance video is the main objective of the work presented in this paper. The method developed consist of various stages of suspicious frame detection, and verifying the frame for suspicious activity related analysis of human movements within obtained set of suspicious frames. The method consist of GLCM feature extraction which constitutes the features such as energy, prominence, contrast, entropy, homogeneity type of features and matching using Euclidian distance along with descriptor features obtained by using Harris corner features and cosine similarity index estimation. The successful suspicious activity detection rate is analyzed which shows better performance and time saving method while analyzing large surveillance video dataset.

Intelligent Video Surveillance using Deep Learning

Now days, Big data applications are having most of the importance and space in industry and research area. Surveillance videos are a major contribution to unstructured big data. The main objective of this paper is to give brief about video analysis using deep learning techniques in order to detect suspicious activities. Our main focus is on applications of deep learning techniques in detection the count, no of involved persons and the activity going on in a crowd considering all conditions [9]. This video analysis helps us to achieve security. Security can be defined in different terms like identification of theft, detecting violence etc. Suspicious Human Activity Detection is simply the process of detection of unusual (abnormal)l human activities . For this we need to convert the video into frames and processing these frames helps us to analyze the persons and their activities. There are two modules in this system first one Object Detection Module and Second one is Activity Detection Module .Object detection module detects whether the object is present or not. After detecting the object the next module is going to check whether the activity is suspicious or not.

Video representation and suspicious event detection using semantic technologies

Storage and analysis of video surveillance data is a significant challenge, requiring video interpretation and event detection in the relevant context. To perform this task, the low-level features including shape, texture, and color information are extracted and represented in symbolic forms. In this work, a methodology is proposed, which extracts the salient features and properties using machine learning techniques and represent this information as Linked Data using a domain ontology that is explicitly tailored for detection of certain activities. An ontology is also developed to include concepts and properties which may be applicable in the domain of surveillance and its applications. The proposed approach is validated with actual implementation and is thus evaluated by recognizing suspicious activity in an open parking space. The suspicious activity detection is formalized through inference rules and SPARQL queries. Eventually, Semantic Web Technology has proven to be a remarkable toolchain to interpret videos, thus opening novel possibilities for video scene representation, and detection of complex events, without any human involvement. The proposed novel approach can thus have representation of frame-level information of a video in structured representation and perform event detection while reducing storage and enhancing semantically-aided retrieval of video data.

Automatic Gun Detection Approach for Video Surveillance

The immense crime rates resulting from using pistols have led governments to seek solutions to deal with such terrorist incidents. These incidents have a negative impact on public security and cause panic among citizens. From this point, facing a pandemic of weapon violence has become an important research topic. One way to reduce this kind of violence is to prevent it via remote detection and to give an appropriate response in a short time. Video surveillance is the process of monitoring the behavior of people and objects. Surveillance systems can be employed in security applications as legal evidence. Moreover, it is used widely in suspicious activity detection applications. Intelligent video surveillance systems (IVSSs) are the use of automatic video analytics to enhance the effectiveness of traditional surveillance systems. With the rapid development in Deep Learning (DL), it is now widely used to address the problems existing in traditional detection techniques. In this article, an approach to detect pistols and guns in video surveillance systems is proposed. The presented approach does not need any invasive tools in the weapon detection process. It uses DL in the classification and the detection processes. The proposed approach enhances the obtained results by applying Transfer Learning (TL). It employs two different DL techniques: AlexNet and GoogLeNet. Experimental results verify the adaptability of detecting different types of pistols and guns. The experiments were conducted on a benchmark gun database called Internet Movie Firearms Database (IMFDB). The results obtained suggest that the proposed approach is promising and outperforms its counterparts.

Towards a Multi-Agent based Network Intrusion Detection System for a Fleet of Drones

The objective of this research work is to propose a new model of intrusion detection system for a fleet of UAVs deployed with an ad hoc communication architecture. The security of a drone fleet is rarely addressed by the scientific community, and most research has focused on routing protocols and battery autonomy, while ignoring the security aspect. The multi-agent paradigm is considered the most adequate and appropriate solution to model an effective intrusion detection system capable of detecting intrusions targeting a drone fleet. Multi-agent systems can perfectly address the security problem of a drone fleet, given the mobility, autonomy, cooperation and distribution characteristics present in the network linking the different nodes of the fleet. The proposed model consists of a set of cooperative, autonomous, communicating, learning and intelligent agents that collaborate with each other to carry out intrusion and suspicious activity detection missions that can target the network of a fleet of drones. Our system is autonomous and can detect known and unknown cyber attacks in real time without the need for human experts, who generally design the signatures of known attacks for conventional intrusion detection systems.

IBaggedFCNet: An Ensemble Framework for Anomaly Detection in Surveillance Videos

The prevalent use of surveillance cameras in public places and advancements in computer vision warrant most sought-after research in the domain of anomalous activity detection. Anomaly detection has shown promising applications for suspicious activity detection. In this paper, we propose a bagging framework IBaggedFCNet that leverages the power of ensembles for robust classification to detect anomalies in videos. Our approach, which investigates state-of-the-art Inception-v3 image classification network, requires no video segmentation prior to feature extraction that can produce unstable segmentation results and cause a high memory footprint. We show improvement empirically on multiple benchmark datasets, most prominently on the UCF-Crime dataset. Moreover, we experiment with different ensemble fusion methods, including static and dynamic techniques, and also prove our single model’s predictive accuracy in localizing anomaly in surveillance videos.

An Efficient Method for Suspicious Activity Detection

This paper contributes on suspicious activity detection from length of video with less complex processing algorithm. The proposed method in this paper is easy to implement and robust enough to monitor different suspicious activities such as sudden seating, standing up, hiding from midway path, entry from midway. The suspicious frame detection is a novel approach and then confirmation is done using SURF based descriptor matching for speedy processing requirements. The results obtained in terms of tracking window and detection capability are satisfactory.

A review of supervised and unsupervised machine learning techniques for suspicious behavior recognition in intelligent surveillance system

There is a strong demand of smart vision based surveillance system owing to the increase in crime at a frightening rate at various public places like Banks, Airport, Shopping malls and its application in human activity recognition ranges from patient fall detection, irregular pattern recognition or Human computer Interaction. As the crime increases at a disturbing rate, public security violations and high cost of security personals have motivated the author to do the strategic survey of existing vision and image processing based techniques in the past literature. The paper begins with discussing the common approach towards suspicious activity detection and recognition followed by summarizing the supervised and unsupervised machine learning methodologies mainly based on SVM, HMM and ANN classifiers, which were adopted by the researchers previously varying from single human behavior modeling to crowded scenes. Next, this paper discusses system model for human’s normal and abnormal activities recognition along with various feature selectors and detectors used in previous literature. This was followed by conducting a review of benchmark researches which covered a comprehensive state of art methodologies in the related fields, key points owned, feature learning and applications. At last experimental aspects of various papers have been discussed with essential performance matrices like accuracy along with the major issues, common problems, challenges and future scope in the related field.

A TIME FREQUENCY BASED SUSPICIOUS ACTIVITY DETECTION FOR ANTIMONEY LAUNDERING

Money laundering is the crucial mechanism utilized by criminals to inject proceeds of crime into the financial system. The primary responsibility of the detection of suspicious activity related to money laundering is with the financial institutions. Most of the current systems in these institutions are rule-based and ineffective (over 90 % false positives). The available data science-based anti-money laundering (AML) models to replace the existing rule-based systems work on customer relationship management (CRM) features and time characteristics of transaction behaviour. Due to thousands of possible account features, customer features, and their combinations, it is challenging to perform feature engineering to achieve reasonable accuracy. Aiming to improve the detection performance of suspicious transaction monitoring systems for AML systems, in this article, we introduce a novel feature set based on time-frequency analysis, that uses 2-D representations of financial transactions. Random forest is utilized as a machine learning method, and simulated annealing is adopted for hyperparameter tuning. The designed algorithm is tested on real banking data, proving the results’ efficacy in practically relevant environments. It is shown that the time-frequency characteristics are discriminatory features for suspicious and non-suspicious entities. Therefore, these features substantially improve the area under curve results (over 1%) of the existing data science-based transaction monitoring systems. Using time-frequency features alone, a false positive rate of 14.9% has been achieved, with an F-score of 59.05%. When combined with transaction and CRM features, the false positive rate is 11.85%, and the F-Score is improved to 74.06%.

A soft clustering technique with layered feature extraction for social image mining

Social image mining is beneficial to accomplish tasks like event detection, suspicious activity detection, prediction of future trends, identification of mentally depressed people, etc. To carry out social image mining, data mining techniques need to be used. Clustering is one of the most important tasks of data mining which is able to deal with the unlabelled data. But, less number of clustering approaches are having ability to handle the uncertain image data. Thus, in this paper we proposed a soft clustering algorithm named as ROugh Mean Shift clustering (ROMS) with layered feature extraction model for social images. Effectiveness of the rough set theory and mean shift concepts are incorporated in this algorithm. It makes the ROMS to deal with the vagueness and the automatic determination of cluster numbers in given data. Proposed method is experimented on three datasets- synthetic, standard and real-world datasets and compared with existing techniques. Experimental results show that ROMS performs better as compared to other techniques.

Intelligent Information System for Suspicious Human Activity Detection in Day and Night

The detection of human beings in a camera attracts more attention because of its wide range of applications such as abnormal event detection, person counting in a dense crowd, person identification, fall detection for care to elderly people, etc. Over the time, various techniques have evolved to enhance the visual information. This article presents a novel 3-D intelligent information system for identifying abnormal human activity using background subtraction, rectification, morphology, neural networks and depth estimation with a thermal camera and a pair of hand held Universal Serial Bus (USB) camera to visualize un-calibrated images. The proposed system detects strongest points using Speed-Up Robust Features (SURF). The Sum of Absolute Difference (SAD) algorithm match the strongest points detected by SURF. 3-D object model and image stitching from image sequences are carried out in the proposed work. A series of images captured from different cameras are stitched into a geometrically consistent mosaic either horizontally/vertically based on the image acquisition. 3-D image and depth estimation of un-calibrated stereo images are acquired using rectification and disparity. The background is separated from the scene using threshold approach. Features are extracted using morphological operators in order to get the skeleton. Junction points and end points of the skeleton image are obtained from the skeleton. Data set of abnormal human activity is created using supervised learning such as neural network with a thermal camera and a pair of webcam. The feature vector of an activity is compared with already created data set, if a match occurs the classifier detects abnormal human activity. Additionally the proposed algorithm performs depth estimation to measure real time distance of objects dynamically. The system use thermal camera, Intel computing stick, converter, video graphics array (VGA) to high-definition multimedia interface (HDMI) and webcams. The proposed novel intelligent information system gives 94% maximum accuracy and 89% minimum accuracy for different activities, thus it effectively detects suspicious activity during day and night.