Statistical Features-Based Violence Detection in Surveillance Videos

Kumar, Deepak; Vignesh, L. K. P.; Srivathsan, G.; Roshan, S.; Chandrakala, S.

doi:10.1007/978-981-15-1451-7_21

Cited by 10 publications

(9 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Datasets: The Hockey Fight [23] and Movies Dataset [24] are adopted for training and evaluating the proposed AMCNN. The Hockey Fight includes 1000 video clips, which consist of an equal number of violent and non-violent behaviors.…”

Section: Datasets and Experiments Settingmentioning

confidence: 99%

Attention Inspiring Receptive-Fields Multi-Task Network via Self- supervised Learning for Violence Recognition

Song

2023

Preprint

View full text Add to dashboard Cite

Generally, a large amount of training data is essential to train deep learning model for obtaining more accurate detection performance in computer vision domain. However, to collect and annotate datasets will lead to extensive cost. In this letter, we propose a self-supervised auxiliary task to learn general videos features without adding any human-annotated labels, aiming at improving the performance of violence recognition. Firstly, we propose a violence recognition method based on convolutional neural network with self-supervised auxiliary task, which can learn visual feature for improving down-stream task (recognizing violence). Secondly, we establish a balance-weighting scheme to solve the crucial problem of balancing the self-supervised auxiliary task and violence recognition task. Thirdly, we develop an attention receptive-field module, indicating that the proper use of the spatial attention mechanism can effectively expand the receptive fields of the module, further improving semantically meaningful representation of the network. To evaluate the proposed method, two benchmark datasets have been used, and better performance is shown by the experimental results comparing with other state-of-the-art methods.

show abstract

Section: Datasets and Experiments Settingmentioning

confidence: 99%

Attention Inspiring Receptive-Fields Multi-Task Network via Self- supervised Learning for Violence Recognition

Song

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…In [29], a new deep NeuralNet approach has been presented for the task of Violence Detection by extraction of motion features in RGB Dynamic Images (DI). Deepak et al [30] introduce a novel statistical feature descriptor for detecting violent human actions in real-time surveillance videos. Ehsan et al [31] presented a novel Vi-Net structure dependent upon the deep CNN for detecting activities with abnormal velocity.…”

Section: Related Workmentioning

confidence: 99%

Tuna Swarm Algorithm With Deep Learning Enabled Violence Detection in Smart Video Surveillance Systems

Aldehim,

Asiri,

Aljebreen

et al. 2023

IEEE Access

View full text Add to dashboard Cite

In smart video surveillance systems, violence detection becomes challenging to ensure public safety and security. With the proliferation of surveillance cameras in public areas, there is an increasing need for automated algorithms that can accurately and efficiently detect violent behavior in real time. This article presents a Tuna Swarm Optimization with Deep Learning Enabled Violence Detection (TSODL-VD) technique to classify violent actions in surveillance videos. The TSODL-VD technique enables the recognition of violence and can be a measure to avoid chaotic situations. In the presented TSODL-VD technique, the residual-DenseNet model is applied for feature vector generation from the input video frames and then passed into the stacked autoencoder (SAE) classifier. The SAE model is enforced to recognize the events into violence and non-violence events. To improve the violence detection effectiveness of the TSODL-VD procedure, the TSO protocol is utilized as a hyperparameter optimizer for the residual-DenseNet model. The performance validation of the TSODL-VD procedure has experimented on a benchmark violence dataset. The experimental results demonstrate that the TSODL-VD technique accomplishes precise and rapid detection outcomes over the recent state-of-the-art approaches.

show abstract

“…The classification is then done using a machine learning classifier that has been trained. Deepak et al [15] detect and distinguish aggressive actions in crowded scenes. To extract motion information from video frames, Spatio-temporal autocorrelation of gradient was used as a feature extraction method.…”

Section: Handcrafted Based Modelsmentioning

confidence: 99%

“…Various classifiers have been employed in both supervised and unsupervised settings. According to the papers reviewed, SVM was the most widely used supervised classifier [15] [29] [30], whereas CNN was the most widely used unsupervised classifier [31] [22] [24]. Anomaly representation has been applied in both categories using distinct features.…”

Section: Classificationmentioning

confidence: 99%

See 1 more Smart Citation

Unusual Activity Detection in Surveillance Video Scene: Review

Mahdi

Mohammed²,

Jafer

2021

J. Al-Qadisiyah Comp. Sci. Math.

View full text Add to dashboard Cite

Abnormal activity may indicate threats and risks to others. An anomaly can be defined as something that deviates from what is expected, common, or normal. Because it is difficult to continuously monitor public spaces, intelligent video surveillance is necessary, which can monitor human actions in real-time and categorize them as ordinary or exceptional, as well as create an alarm. Human activities in public and sensitive regions such as bus stations, airports, railway stations, malls, banks, universities, car parks, roads, and other regions can be monitored using visual surveillance to prevent crime, theft, terrorism, vandalism, accidents, and other suspicious activities. This makes video surveillance a key to increasing public security. The main objective of event discovery is to discover the occurrence of events and categorize them into normal or abnormal actions. This discovery requires identifying and tracking objects and then learning what is going around those tracked objects. Recent research is based on one of two technologies: handcrafted features and deep learning models. Handmade features are based on extracting low-level features, and their strength is based on selecting the best features, that produce the best results. After the success of deep learning techniques for classifying images, the researchers examined the ability of deep learning techniques to detect, which bypasses the manual step of feature extraction and works directly with images. This paper presents a survey of both handmade and deep learning models to detect abnormal events.

show abstract

Statistical Features-Based Violence Detection in Surveillance Videos

Cited by 10 publications

References 5 publications

Attention Inspiring Receptive-Fields Multi-Task Network via Self- supervised Learning for Violence Recognition

Attention Inspiring Receptive-Fields Multi-Task Network via Self- supervised Learning for Violence Recognition

Tuna Swarm Algorithm With Deep Learning Enabled Violence Detection in Smart Video Surveillance Systems

Unusual Activity Detection in Surveillance Video Scene: Review

Contact Info

Product

Resources

About