Review on Deep Learning Approaches for Anomaly Event Detection in Video Surveillance

Electronics, Journal Year: 2022, Volume and Issue: 12(1), P. 29 - 29

Published: Dec. 22, 2022

In the last few years, due to continuous advancement of technology, human behavior detection and recognition have become important scientific research in field computer vision (CV). However, one most challenging problems CV is anomaly (AD) because complex environment difficulty extracting a particular feature that correlates with event. As number cameras monitoring given area increases, it will vital systems capable learning from vast amounts available data identify any potential suspicious behavior. Then, introduction deep (DL) has brought new development directions for AD. particular, DL models such as convolution neural networks (CNNs) recurrent (RNNs) achieved excellent performance dealing AD tasks, well other domains like image classification, object detection, speech processing. this review, we aim present comprehensive overview those methods using address problem. Firstly, different classifications anomalies are introduced, then architectures used video discussed analyzed, respectively. The revised contributions been categorized by network type, architecture model, datasets, metrics evaluate these methodologies. Moreover, several applications discussed. Finally, outlined challenges future further field.

Language: Английский

---

Novel Deep Feature Fusion Framework for Multi-Scenario Violence Detection

Computers, Journal Year: 2023, Volume and Issue: 12(9), P. 175 - 175

Published: Sept. 5, 2023

Detecting violence in various scenarios is a difficult task that requires high degree of generalisation. This includes fights different environments such as schools, streets, and football stadiums. However, most current research on detection focuses single scenario, limiting its ability to generalise across multiple scenarios. To tackle this issue, paper offers new multi-scenario framework operates two environments: fighting locations rugby has three main steps. Firstly, it uses transfer learning by employing pre-trained models from the ImageNet dataset: Xception, Inception, InceptionResNet. approach enhances generalisation prevents overfitting, these have already learned valuable features large diverse dataset. Secondly, combines extracted through feature fusion, which improves representation performance. Lastly, concatenation step first scenario with second train machine classifier, enabling classifier both highly flexible, can incorporate without requiring training scratch additional The Fusion model, incorporates fusion models, obtained an accuracy 97.66% RLVS dataset 92.89% Hockey Concatenation model accomplished 97.64% 92.41% datasets just classifier. allows for classification violent within Furthermore, not limited be adapted tasks.

Language: Английский

---

Hiding Information in Digital Images Using LSB Steganography Technique

International Journal of Interactive Mobile Technologies (iJIM), Journal Year: 2023, Volume and Issue: 17(07), P. 167 - 178

Published: April 5, 2023

The highest way to protect data from intruder and unauthorized persons has become a major issue. This matter led the development of many techniques for security, such as Steganography, Cryptography, Watermarking disguise data. paper proposes an image steganography method using Least Significant Bits (LSB) technique XOR operator secret key, through which key is transformed into one-dimensional bit stream array, then these bits are XORed with image. Multiple experiments have been performed embed color grayscale images inside cover media. In this work, LSB ideal in two ways: firstly, only least significant one-bit (1bit) each byte will store embedded data, named (1-LSB). Secondly, four right half-byte (4 bits) (4-LSB). Subjective objective analyzes were process. subjective analysis responsible both HVS histogram, whereas involved PSNR MSE metrics.

Language: Английский

---

Effect of Changing Targeted Layers of the Deep Dream Technique Using VGG-16 Model

International Journal of Online and Biomedical Engineering (iJOE), Journal Year: 2023, Volume and Issue: 19(03), P. 34 - 47

Published: March 14, 2023

The deep dream is one of the most recent techniques in learning. It used many applications, such as decorating and modifying images with motifs simulating patients' hallucinations. This study presents a model that generates using convolutional neural network (CNN). Firstly, we survey layers each block network, then choose required layers, extract their features to maximize it. process repeats several iterations needed, computes total loss, extracts final images. We apply this operation on different two times; former low-level latter high-level layers. results applying are different, where resulting image from clearer than those Also, loss ranges between 31.1435 31.1435, while upper 20.0704 32.1625.

Language: Английский

---

Employing the Concept of Stacking Ensemble Learning to Generate Deep Dream Images Using Multiple CNN Variants

Intelligent Systems with Applications, Journal Year: 2025, Volume and Issue: unknown, P. 200488 - 200488

Published: Jan. 1, 2025

Language: Английский

---

A Scalable and Generalised Deep Learning Framework for Anomaly Detection in Surveillance Videos

International Journal of Intelligent Systems, Journal Year: 2025, Volume and Issue: 2025(1)

Published: Jan. 1, 2025

Anomaly detection in videos is challenging due to the complexity, noise, and diverse nature of activities such as violence, shoplifting, vandalism. While deep learning (DL) has shown excellent performance this area, existing approaches have struggled apply DL models across different anomaly tasks without extensive retraining. This repeated retraining time‐consuming, computationally intensive, unfair. To address limitation, a new framework introduced study, consisting three key components: transfer enhance feature generalization, model fusion improve representation, multitask classification generalize classifier multiple training from scratch when task introduced. The framework’s main advantage its ability requiring for each task. Empirical evaluations demonstrate effectiveness, achieving an accuracy 97.99% on RLVS (violence detection), 83.59% UCF dataset (shoplifting 88.37% both datasets using single Additionally, tested unseen dataset, achieved 87.25% 79.39% violence shoplifting datasets, respectively. study also utilises two explainability tools identify potential biases, ensuring robustness fairness. research represents first successful resolution generalization issue detection, marking significant advancement field.

Language: Английский

---

A Systematic Review of Deep Dream

Iraqi Journal of Computer Communication Control and System Engineering, Journal Year: 2023, Volume and Issue: unknown, P. 192 - 209

Published: June 29, 2023

Deep Dream (DD) is a new technology that works as creative image-editing approach by employing the representations of CNN to produce dreams-like images taking benefits both and Inception build dream through layer-by-layer implementation. As days go by, DD becomes widely used in artificial intelligence (AI) fields. This paper first systematic review DD. We focused on definition, importance, background, applications Natural language processing (NLP), images, videos, audio are main fields which applied. also discussed concepts DD, like transfer learning Inception. addressed contributions, databases, techniques have been models, limitations, evaluation metrics for each one included research papers. Finally, some interesting recommendations listed serve researchers future. Index Terms— dream, deep CNN, gradient ascent, Inception, style transfer.

Language: Английский

---

Abnormal Behavior Detection in Video Surveillance Using Inception-v3 Transfer Learning Approaches

Iraqi Journal of Computer Communication Control and System Engineering, Journal Year: 2023, Volume and Issue: unknown, P. 210 - 221

Published: June 29, 2023

The use of video surveillance systems has increased due to security concerns and their relatively low cost. Researchers are working create intelligent Closed Circuit Television (CCTV) cameras that can automatically analyze behavior in real-time detect anomalous behaviors prevent dangerous accidents. Deep Learning (DL) approaches, particularly Convolutional Neural Networks (CNNs), have shown outstanding results analysis anomaly detection. This research paper focused on using Inception-v3 transfer learning approaches improve the accuracy efficiency abnormal detection surveillance. network is used classify keyframes a as normal or by utilizing both pre-training fine-tuning extract features from input data develop new classifier. UCF-Crime dataset train evaluate proposed models. performance models was evaluated accuracy, recall, precision, F1 score. fine-tuned model achieved 88.0%, 89.24%, 85.83%, 87.50% for these measures, respectively. In contrast, pre-trained obtained 86.2%, 86.43%, 84.62%, 85.52%, These demonstrate architecture effectively videos, weights layers further model's performance. Index Terms— Abnormal detection, Video surveillance, learning, Transfer InceptionV3.

Language: Английский

---

A Hybrid Artistic Model Using Deepy-Dream Model and Multiple Convolutional Neural Networks Architectures

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 101443 - 101459

Published: Jan. 1, 2023

The significant increase in drug abuse cases prompts developers to investigate techniques that mimic the hallucinations imagined by addicts and abusers, addition increasing demand for use of decorative images resulting from computer technologies. This research uses Deep Dream Neural Style Transfer technologies solve this problem. Despite significance researches on technology, there are several limitations existing studies, including image quality evaluation metrics. We have successfully addressed these issues improving diversifying types generated images. enhancement allows more effective simulating hallucinated Moreover, high-quality can be saved dataset enlargement, like augmentation process. Our proposed deepy-dream model combines features five convolutional neural network architectures: VGG16, VGG19, Inception v3, Inception-ResNet-v2, Xception. Additionally, we generate implementing each architecture as a separate model. employed autoencoder another method. To evaluate performance our models, utilize normalized cross-correlation structural similarity indexes values obtained those two measures 0.1863 0.0856, respectively, indicating performance. When considering content image, metrics yield 0.8119 0.3097, respectively. Whiefor style corresponding measure 0.0007 0.0073,

Language: Английский

---

Generating Various Deep Dream Images Through Maximizing the Loss Function of Particular Layers Using Inception-v3 and Inception-ResNet-V2 Models

Iraqi Journal of Science, Journal Year: 2024, Volume and Issue: unknown, P. 3468 - 3483

Published: June 30, 2024

Recently, Deep Learning (DL) has been used in a new technology known as the Dream (DD) to produce images that resemble dreams. It is utilized mimic hallucinations drug users or people with schizophrenia experience. Additionally, DD sometimes incorporated into decoration. This study produces using two deep-CNN model architectures (Inception-ResNet-V2 and Inception-v3). starts by choosing particular layers each (from both lower upper layers) maximize their activation function, then detect several iterations. In iteration, gradient computed compute loss present resulting images. Finally, total presented, final deep dream image visualized. The output of models different, even for same there are some variations, layers' values Inception-v3 significantly higher comparison levels' values. case Inception-ResNet-V2, convergent.

Language: Английский

---