Campus Abnormal Behavior Recognition With Temporal Segment Transformers

Citizen insecurity in underdeveloped third world countries is aggravated by poor management of arms control and illegal trafficking, which requires information technology solutions in intelligent video surveillance systems for the detection of lethal weapons. The literature review highlights the need for an intelligent video surveillance system to combat high crime, using fog computing, which processes data in real time for the detection of weapons and other crimes. Furthermore, at an international level, solutions based on artificial intelligence and deep learning are being implemented for object recognition and weapons detection. Therefore, this paper describes the design of an intelligent video surveillance system based on artificial vision, fog and edge computing to detect lethal weapons in domestic environments, performing weapon classification and data transmission to police centers. The intelligent video surveillance system allows detecting lethal weapons and operates in three stages: an edge node with a Raspberry Pi 4; a detection algorithm based on a convolutional neural network with YOLOv5; and streaming tagged images to a security unit via WhatsApp. The main conclusion is that the system achieved a precision greater than 0.85 and a quick and efficient response in sending alerts, representing a scalable and effective solution against home burglary.

Public surveillance monitoring is rapidly finding its way into Intelligent Surveillance System. Street crime is increasing in recent years, which has demanded more reliable and intelligent public surveillance system. In this paper, the ability and the accuracy of an Adaptive Neuro-Fuzzy Inference System (ANFIS) was investigated for the classification of moving objects for street scene applications. The goal of this paper is to classify the moving objects prior to its communal attributes that emphasize on three major processes which are object detection, discriminative feature extraction, and classification of the target. The intended surveillance application would focus on street scene, therefore the target classes of interest are pedestrian, motorcyclist, and car. The adaptive network based on Neuro-fuzzy was independently developed for three output parameters, each of which constitute of three inputs and 27 Sugeno-rules. Extensive experimentation on significant features has been performed and the evaluation performance analysis has been quantitatively conducted on three street scene dataset, which differ in terms of background complexity. Experimental results over a public dataset and our own dataset demonstrate that the proposed technique achieves the performance of 93.1% correct classification for street scene with moving objects, with compared to the solely approaches of neural network or fuzzy.

In recent years, human motion recognition in computer vision has become a hot research direction in this field. Based on 2D human motion recognition technology, real-time extraction of motion features from 2D planes is used to recognize human movements. This method can only learn the position contour and color information of the image. It cannot directly reflect the motion situation, which results in low recognition accuracy and efficiency. In response to this issue, this study proposes a combination method of motion recognition and 3D pose estimation to recognize and classify basketball movements. First, the 2D skeleton model is obtained by extracting the feature information in the video action, which is converted into a 3D model, and the model is replaced by the time-space convolutional network to establish a human action recognition model. The experiment showed that when the number of iterations reached 6, the accuracy of the spatio-temporal graph convolutional network algorithm model reached 92%. Comparing the accuracy rates of different algorithm models, the average accuracy rates of convolutional neural network, long short memory network, graph convolution, long short model of action recognition and graph convolution model of action recognition were 61.6%, 65.4%, 72.5%, 76.8% and 90.3% respectively. The results show that the proposed 3D recognition algorithm can accurately recognize different basketball movements. This study can provide reference for basketball coaches and athletes in basketball training.

Computer vision technologies have greatly improved in the last few years. Many problems have been solved using deep learning merged with more computational power. Action recognition is one of society's problems that must be addressed. Human Action Recognition (HAR) may be adopted for intelligent video surveillance systems, and the government may use the same for monitoring crimes and security purposes. This paper proposes a deep learning-based HAR model, i.e., a 3-dimensional Convolutional Network with multiplicative LSTM. The suggested model makes it easier to comprehend the tasks that an individual or team of individuals completes. The four-phase proposed model consists of a 3D Convolutional neural network (3DCNN) combined with an LSTM multiplicative recurrent network and Yolov6 for real-time object detection. The four stages of the proposed model are data fusion, feature extraction, object identification, and skeleton articulation approaches. The NTU-RGB-D, KITTI, NTU-RGB-D 120, UCF 101, and Fused datasets are some used to train the model. The suggested model surpasses other cutting-edge models by reaching an accuracy of 98.23%, 97.65%, 98.76%, 95.45%, and 97.65% on the abovementioned datasets. Other state-of-the-art (SOTA) methods compared in this study are traditional CNN, Yolov6, and CNN with BiLSTM. The results verify that actions are classified more accurately by the proposed model that combines all these techniques compared to existing ones.

The aim of the work is to develop algorithms for analyzing video data in real time based on computer vision methods, as well as deep learning technologies for artificial neural networks for abnormal human behavior recognition near critical facilities using ATMs as an example. The article provides an overview of the initial research aimed at the choice of data capture devices, neural network architecture, software implementation and selection of experimental conditions (distance and illumination). Static and dynamic hand gestures were used as object’s movements. Experimental results show that using the Intel RealSense D435 Depth Camera provides more accurate dynamic gesture recognition under different experimental conditions.

Human action recognition has been a popular issue recently because it can be applied in many applications such as intelligent surveillance systems, human-robot interaction, and autonomous vehicle control. Human action recognition using RGB video is a challenging task because the learning of actions is easily affected by the cluttered background. To cope with this problem, the proposed method estimates 3D human poses first which can help remove the cluttered background and focus on the human body. In addition to the human poses, the proposed method also utilizes appearance features nearby the predicted joints to make our action prediction context-aware. Instead of using 3D convolutional neural networks as many action recognition approaches did, the proposed method uses a two-stream architecture that aggregates the results from skeleton-based and appearance-based approaches to do action recognition. Experimental results show that the proposed method achieved state-of-the-art performance on NTU RGB+D which is a large-scale dataset for human action recognition.

In this paper, a new approach of learning features based on convolutional neural networks for pedestrian detection in far infrared images is presented. Unlike traditional recognition systems which use hand-designed features like SIFT or HOG, our convolutional networks architecture learns new features and representations more appropriate to the classification task in infrared images. Another pedestrian detector based on logistic regression is designed and compared to convolutional networks based classifier. Our system built over non-visible range sensor may have an important role in next generation robotics, especially in perception, advanced driver assistant systems (ADAS) and intelligent surveillance systems.

Enormous advance has proven throughout the years in the area of traffic surveillance by the growth of intelligent traffic video surveillance system. In the current work, through the traffic videos, the traffic video surveillance automatically keyed out the vehicles like ambulance and trucks, which in turn assisted us in directing the vehicles at the time of emergency. Nevertheless, it doesn’t provide us a vital solution for the regulating the traffic. Moreover, this idea just identifies the vehicles, but it couldn’t notice the accidents expeditiously. Therefore in the proposed work, expeditious traffic video surveillance and monitoring system are presented along with dynamic traffic signal control and accident detection mechanism. Hybrid median filter has been utilized at the beginning for pre-processing of traffic videos, and to remove the noise. Hybrid support vector machine (SVM with extended Kalman filter) has been utilized to chase the vehicles. Next, the histogram of flow gradient features are drew-out to categories the vehicles. According to the traffic density and through video files, vehicles are computed, and then for emergency vehicles, the traffic signal gets switched dynamically. To realize the arrival of ambulances, the cameras have been set to catch traffic videos minimum at 500 m of the signal and deep learning neural networks has been employed. Hence dynamic signal control has been incorporated expeditiously. Likewise, multinomial logistic regression has been utilized in real-time live streaming videos, to identify the accidents correctly. The observational solution shows that the proposed intelligent traffic video surveillance system render expeditious dynamic control of traffic signals and it raises the identification of accidents correctly.

This paper presents a VLSI architecture of real-time object detection design for the intelligent video surveillance systems. In order to achieve the targets of high performance, low cost, and high accuracy, an efficient block-based background subtraction (BBS) algorithm had been created for VLSI implementation. It included a background model, a luminance generator, and a block difference model. The warning signals in the intelligent video surveillance systems will be triggered once more than a threshold number of blocks are marked as object discovered blocks. The VLSI architecture of this work contained 4.6-K gate counts and consumed 2.05 mW when it operated at 100 MHz processing rate by using a TSMC 0.18 um CMOS process. Compared with previous low-complexity designs, this work had the benefits of lower cost, higher performance and higher accuracy.

This paper proposes a new intelligent video surveillance system based on Android phone. By using Android phone, this system can be deployed easily and quickly. Compare with the traditional surveillance system, this new one uses Android phone to replace the large and expensive surveillance cameras. In addition, the functions of back-end equipment of the traditional system have been moved to the front-end equipment. With a modified invasion detection algorithm, the front-end equipment becomes more intelligent. This schema can significantly increase the mobility and flexibility of the entire system. Therefore this system will have a better performance dealing with emergencies in different surveillance scenarios. Moreover, this state-of-art intelligent video surveillance system not only brings a new method for surveillance, but also advances the development of surveillance techniques.

Falls are a growing social problem and have become a hot topic in healthcare. Thanks to recent advances in deep convolutional neural networks, the accuracy of video-based fall detection has been greatly improved. However, these methods are affected by illumination and complex backgrounds. Video angles and other influencing factors reduce the accuracy and generalization ability of these methods. In this paper, a video-based human fall detection method is proposed. First, a 2D joint point sequence in the video is extracted using a pose estimator, and then a 2D joint point pose sequence is extracted. It is elevated to a 3D joint point pose sequence and then recognized whether it is a fall action by our improved multi-scale unified spatial-temporal graph convolutional network (MS-G3D). The system proves its effectiveness and robustness in the field of action recognition, achieving 99.84% accuracy on the large benchmark action recognition dataset NTU RGB+D, and 95.72% accuracy on the LE2I fall dataset.

In recent years, activity recognition in smart homes is an active research area. It is an important problem of Human Computer Interaction (HCI), and has many applications in HCI, such as assistive living and healthcare. Recognition of users’ common behaviors allows an environment to provide personalized service. Unlike activity recognition in computer vision which uses cameras, it studies activities by embedded sensors in smart homes. In this paper, we propose a method to extract latent features from sensor data by Beta Process Hidden Markov Model (BP-HMM). The contributions of our method are twofold: 1, we extend BP-HMM by dependent Beta process, and integrate state constraints of sensors into the sampling process of BP-HMM. 2, we extract latent features automatically by our dependent BP-HMM, and train a structural support vector machine (SVM) by these features in a supervised way for activity recognition. To evaluate the proposed method, we performed experiments on the real-world smart home datasets. Our results suggest that extracting latent features from sensor data leads to good performance for activity recognition.

Moving target detection and tracking, recognition, behaviors analysis are the key issues in the intelligent visual surveillance system (IVSS). The challenge is how to process the real-time video stream in an effective way in case that we could find the interested objects for analysis. However, the traditional video surveillance technology often does not meet the needs of real-time key frame recognition for the on-line intelligent video monitoring system. In our paper, we adopt the state-of-the-art Faster R-CNN [7] that takes advantages of convolutional neural networks into our real-time target recognition system - Deep Intelligent Visual Surveillance (DIVS). The key aspects of our DIVS are consisted of four parts: (i) Getting the real-time video images from remote cameras; (ii) Processing the data with the deep learning framework caffe [23] built for Faster R-CNN; (iii) Storing the valuable data with MySQL; (iv) Data presentation on the website. Experiments based on our system validated the effectiveness, stability and accuracy of our proposed solutions.

With the development of video surveillance gradually from the traditional security surveillance to intelligent surveillance, in order to find an intelligent surveillance model to facilitate the implementation and promotion, we present a method, performing the moving object detection by using OpenCV based on DirectShow framework, which will make intelligent surveillance come true consequently. In this paper, it will first introduce a real-time priority method and some improvement of moving object detection. Secondly, the filter components that have moving object detection function will be designed by the combination method of DirectShow and OpenCV encapsulation technologies. Finally, it will present the design method and its advantages of building intelligent surveillance system on the basis of the above description.

In last few decades, technological revolution has accelerated the deployment of large scale surveillance systems on almost all public places such as malls, hospitals, airports, railways, bus stations, roads, etc. These intelligent surveillance systems can play crucial role in governance of situations, collective security and safety, mitigating as well as prevention of adversaries. With gradual increase in multi camera surveillance systems enclosing multi angle views of same as well as different scenes has increased complexity of monitoring the systems by manual inspection. Abnormalities also known as anomalies or outliers are inevitable part of the existence and presumed to be rare in occurrence. Manual monitoring of such abnormalities is susceptible to errors and limited by human capabilities such as inattention and tiresome. Hence in the field of computer vision, automated abnormal activity recognition (AAR) from surveillance systems is emerging research area. The intent of this research is to shed a light on recent innovations and developments that have made a mark in abnormal activity recognition (AAR) involving deep learning. This paper also includes conventional categorization of anomalies based on different perspectives which can provide better understanding to young researchers. Though recent developments still poses many real time challenges in automatic abnormal activity recognition, some of them are enclosed in this paper.