Automated Surveillance System Research Articles

Validation of automated surveillance of healthcare-associated infections using electronic screening algorithms

Background: Surveillance of healthcare-associated infection (HAI) is the basis of infection prevention programs. However, manual review of medical records is a labor-intensive and time-consuming process. We evaluated the diagnostic performance of automated surveillance of HAI using electronic screening algorithms. Methods: Between April and June 2022, we conducted surveillance of HAI manually and automatically using electronic screening algorithm on 75 units (general medical and surgical wards and ICUs) in a 2,700-bed, tertiary-care hospital in South Korea. Algorithms for surveillance of HAI were developed accordance with NHSN surveillance definitions (Fig. 1). Catheter-associated urinary tract infections (CAUTIs) were automatically detected when eligible pathogen and fever (>38°C) were matched within infection window period. Other specific types of infection were automatically classified based on laboratory results that met NHSN criteria. After the algorithm showed possible cases that met laboratory-confirmed bloodstream infection (LCBI) criteria, we excluded secondary BSIs using the automatic surveillance algorithm. We analyzed sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) for the automated surveillance system compared to manual surveillance. Results: An algorithm for detecting CAUTI showed 98.7% sensitivity (78 of 79), 100.0% specificity (2,443 of 2,443), 100.0% PPV (78 of 78), and 100.0% NPV (2,443 of 2,444). For CLABSI, the algorithm had 97.3% sensitivity (214 of 220), 98.3% specificity (5,759 of 5,861), 67.7% PPV (214 of 316), and 99.9% NPV (5,759 of 5,765). In total, 102 cases of possible CLABSI were identified by the algorithm, and 76 (74.5%) were eventually diagnosed as secondary BSIs. Also, by chart review, 20 BSIs (19.6%) were present on arrival in ER (ER-POA). In 4 cases (3.9%), an original pathogen reoccurred in a repeated infection timeframe (RIT), and 2 cases (2%) were mucosal barrier injury-LCBI (MBI-LCBI). When we additionally performed manual surveillance for intra-abdominal infection secondary BSI, ER-POA, and assigning pathogen to original BSI in RIT, PPV increased to 87.7% (214 of 244). Conclusions: Algorithm for automated surveillance of CAUTI had good performance; however, automated surveillance of CLABSI was suboptimal. More elaborate screening algorithm for diagnosis CLABSI is needed, and further studies are needed to determine whether an automated surveillance system can reduce workload for surveillance of HAI.Disclosures: None

Clusters of healthcare-associated SARS-CoV-2 infections in Norwegian hospitals detected by a fully automatic register-based surveillance system

Notifications to the Norwegian Institute of Public Health of outbreaks in Norwegian healthcare institutions are mandatory by law, but under-reporting is suspected due to failure to identify clusters, or because of human or system-based factors. This study aimed to establish and describe a fully automatic, register-based surveillance system to identify clusters of healthcare-associated infections (HAIs) of SARS-CoV-2 in hospitals and compare these with outbreaks notified through the mandated outbreak system Vesuv. We used linked data from the emergency preparedness register Beredt C19, based on the Norwegian Patient Registry and the Norwegian Surveillance System for Communicable Diseases. We tested two different algorithms for HAI clusters, described their size and compared them with outbreaks notified through Vesuv. A total of 5033 patients were registered with an indeterminate, probable, or definite HAI. Depending on the algorithm, our system detected 44 or 36 of the 56 officially notified outbreaks. Both algorithms detected more clusters then officially reported (301 and 206, respectively). It was possible to use existing data sources to establish a fully automatic surveillance system identifying clusters of SARS-CoV-2. Automatic surveillance can improve preparedness through earlier identification of clusters of HAIs, and by lowering the workloads of infection control specialists in hospitals.

Image Dataset for Neural Network Performance Estimation with Application to Maritime Ports

Automated surveillance systems based on machine learning and computer vision constantly evolve to improve shipping and assist port authorities. The data obtained can be used for port and port property surveillance, traffic density analysis, maritime safety, pollution assessment, etc. However, due to the lack of datasets for video surveillance and ship classification in real maritime zones, there is a need for a reference dataset to compare the obtained results. This paper presents a new dataset for estimating detection and classification performance which provides versatile ship annotations and classifications for passenger ports with a large number of small- to medium-sized ships that were not monitored by the automatic identification system (AIS) and/or the vessel traffic system (VTS). The dataset is considered general for the Mediterranean region since many ports have a similar maritime traffic configuration as the Port of Split, Croatia. The dataset consists of 19,337 high-resolution images with 27,849 manually labeled ship instances classified into 12 categories. The vast majority of the images contain the port and starboard sides of the ships. In addition, the images were acquired in a real maritime zone at different times of the year, day, weather conditions, and sea state conditions.

Vehicle Smoke Synthesis and Attention-Based Deep Approach for Vehicle Smoke Detection

Third world countries are suffering from extreme vehicular air pollution due to dominating number of fossil fuel-driven vehicles on the road. Therefore, in these countries, automatic surveillance systems are in high demand for close monitoring to identify and penalize vehicles emitting excessive smoke. In recent times, deep learning-based computer vision systems are rigorously working on the same. Their accuracy strongly depends on the number of the training images taken under various imaging conditions. However, there are very few publicly available vehicle smoke datasets that could be used for training purposes. To capture on-road videos for the creation of a dataset is another challenging and time-consuming task. To aid and enhance the vehicular smoke monitoring system, in this article, we propose, a holistic dual-level framework for dataset enhancement by smoke generation along with a transformer network for efficient identification. We have created a realistic vehicle smoke generation algorithm using a range of mask patterns and filtering, which helps us to train our deep neural model by generating sufficient synthetic data. We have also proposed a transformer network on the YOLOv5 backbone, which efficiently identifies the smoke region and the smoky vehicle from the image frame simultaneously. We have shown that the lambda-implemented attention-based detection network outperforms the other state-of-the-art techniques on three baseline datasets. Sample demo videos are available at the link <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/srimantacse/VehicleSmoke</uri> .

Perspectives of Patients With Orthopedic Trauma on Fully Automated Digital Physical Activity Measurement at Home: Cross-sectional Survey Study.

The automated digital surveillance of physical activity at home after surgical procedures could facilitate the monitoring of postoperative follow-up, reduce costs, and enhance patients' satisfaction. Data on the willingness of patients with orthopedic trauma to undergo automated home surveillance postoperatively are lacking. The aims of this study were to assess whether patients with orthopedic trauma would be generally willing to use the proposed automated digital home surveillance system and determine what advantages and disadvantages the system could bring with it. Between June 2021 and October 2021, a survey among outpatients with orthopedic trauma who were treated at a European level 1 trauma center was conducted. The only inclusion criterion was an age of at least 16 years. The paper questionnaire first described the possibility of fully automated movement and motion detection (via cameras or sensors) at home without any action required from the patient. The questionnaire then asked for the participants' demographics and presented 6 specific questions on the study topic. In total, we included 201 patients whose mean age was 46.9 (SD 18.6) years. Most of the assessed patients (124/201, 61.7%) were male. Almost half of the patients (83/201, 41.3%) were aged between 30 and 55 years. The most stated occupation was a nine-to-five job (62/199, 30.8%). The majority of the participants (120/201, 59.7%) could imagine using the proposed measurement system, with no significant differences among the genders. An insignificant higher number of younger patients stated that they would use the automated surveillance system. No significant difference was seen among different occupations (P=.41). Significantly more young patients were using smartphones (P=.004) or electronic devices with a camera (P=.008). Less than half of the surveyed patients (95/201, 47.3%) stated that they were using tracking apps. The most stated advantages were fewer physician visits (110/201, 54.7%) and less effort (102/201, 50.7%), whereas the most prevalent disadvantage was the missing physician-patient contact (144/201, 71.6%). Significantly more patients with a part-time job or a nine-to-five job stated that data analysis contributes to medical progress (P=.047). Most of the assessed participants (120/201, 59.7%) stated that they would use the automated digital measurement system to observe their postoperative follow-up and recovery. The proposed system could be used to reduce costs and ease hospital capacity issues. In order to successfully implement such systems, patients' concerns must be addressed, and further studies on the feasibility of these systems are needed.

Robust Wheel Detection for Vehicle Re-Identification.

Vehicle re-identification is a demanding and challenging task in automated surveillance systems. The goal of vehicle re-identification is to associate images of the same vehicle to identify re-occurrences of the same vehicle. Robust re-identification of individual vehicles requires reliable and discriminative features extracted from specific parts of the vehicle. In this work, we construct an efficient and robust wheel detector that precisely locates and selects vehicular wheels from vehicle images. The associated hubcap geometry can hence be utilized to extract fundamental signatures from vehicle images and exploit them for vehicle re-identification. Wheels pattern information can yield additional information about vehicles in questions. To that end, we utilized a vehicle imagery dataset that has thousands of side-view vehicle collected under different illumination conditions and elevation angles. The collected dataset was used for training and testing the wheel detector. Experiments show that our approach could detect vehicular wheels accurately for 99.41% of the vehicles in the dataset.

Real‐time video anomaly detection for smart surveillance

AbstractHuman monitoring of surveillance cameras for anomaly detection may be a monotonous task as it requires constant attention to judge if the captured activities are anomalous or suspicious. This paper exploits background subtraction (BS), convolutional autoencoder, and object detection for a fully automated surveillance system. BS was performed by modelling each pixel as a mixture of Gaussians (MoG) to concatenate only the higher‐order learning in the foreground. Next, the foreground objects are fed to the convolutional autoencoders to filter out abnormal events from normal ones and automatically identify signs of threat and violence in real time. Then, object detection is introduced on the entire scene and the region of interest is highlighted with a bounding box to minimize human intervention in video stream processing. At recognition time, the network generates an alarm for the presence of an anomaly to notify of the identification of potentially suspicious actions. Finally, the complete system is validated upon several benchmark datasets and proved to be robust for complex video anomaly detection. The (AUC) average area under the curve for the frame‐level evaluation for all benchmarks is 94.94%. The best improvement ratio of AUC between the proposed system and state‐of‐the‐art methods is 7.7%.

A new automated national register-based surveillance system for outbreaks in long-term care facilities in Norway detected three times more severe acute respiratory coronavirus virus 2 (SARS-CoV-2) clusters than traditional methods.

To develop and test a new automated surveillance system that can detect, define and characterize infection clusters, including coronavirus disease 2019 (COVID-19), in long-term care facilities (LTCFs) in Norway by combining existing national register data. The numerous outbreaks in LTCFs during the COVID-19 pandemic highlighted the need for accurate and timely outbreak surveillance. As traditional methods were inadequate, we used severe acute respiratory coronavirus virus 2 (SARS-CoV-2) as a model to test automated surveillance. We conducted a nationwide study using data from the Norwegian preparedness register (Beredt C19) and defined the study population as an open cohort from January 2020 to December 2021. We analyzed clusters (≥3 individuals with positive SARS-CoV-2 test ≤14 days) by 4-month periods including cluster size, duration and composition, and residents' mortality associated with clusters. The study population included 173,907 individuals; 78% employees and 22% residents. Clusters were detected in 427 (43%) of 993 LTCFs. The median cluster size was 4-8 individuals (maximum, 50) by 4-month periods, with a median duration of 9-17 days. Employees represented 60%-82% of cases in clusters and were index cases in 60%-90%. In the last 4-month period of 2020, we detected 107 clusters (915 cases) versus 428 clusters (2,998 cases) in the last period of 2021. The 14-day all-cause mortality rate was higher in resident cases from the clusters. Varying the cluster definitions changed the number of clusters. Automated national surveillance for SARS-CoV-2 clusters in LTCFs is possible based on existing data sources and provides near real-time detailed information on size, duration, and composition of clusters. Thus, this system can assist in early outbreak detection and improve surveillance.

Predicting emerging SARS-CoV-2 variants of concern through a One Class dynamic anomaly detection algorithm

ObjectivesThe objective of this study is the implementation of an automatic procedure to weekly detect new SARS-CoV-2 variants and non-neutral variants (variants of concern (VOC) and variants of interest (VOI)).MethodsWe...

An autonomous decision-making framework for gait recognition systems against adversarial attack using reinforcement learning

Gait identification based on Deep Learning (DL) techniques has recently emerged as biometric technology for surveillance. We leveraged the vulnerabilities and decision-making abilities of the DL model in gait-based autonomous surveillance systems when attackers have no access to underlying model gradients/structures using a patch-based black-box adversarial attack with Reinforcement Learning (RL). These automated surveillance systems are secured, blocking the attacker’s access. Therefore, the attack can be conducted in an RL framework where the agent’s goal is determining the optimal image location, causing the model to perform incorrectly when perturbed with random pixels. Furthermore, the proposed adversarial attack presents encouraging results (maximum success rate = 77.59%). Researchers should explore system resilience scenarios (e.g., when attackers have no system access) before using these models in surveillance applications.

Effect of a real-time automatic nosocomial infection surveillance system on hospital-acquired infection prevention and control

BackgroundThe systematic collection of valid data related to hospital-acquired infections (HAIs) is considered effective for nosocomial infection prevention and control programs. New strategies to reduce HAIs have recently fueled the adoption of real-time automatic nosocomial infection surveillance systems (RT-NISSs). Although RT-NISSs have been implemented in some hospitals for several years, the effect of RT-NISS on HAI prevention and control needs to be further explored.MethodsA retrospective, descriptive analysis of inpatients from January 2017 to December 2019 was performed. We collected hospital-acquired infection (HAI) cases and multidrug resistant organism (MDRO) infection cases by traditional surveillance in period 1 (from January 2017 to December 2017), and these cases were collected in period 2 (from January 2018 to December 2018) and period 3 (from January 2019 to December 2019) using a real-time nosocomial infection surveillance system (RT-NISS). The accuracy of MDRO infection surveillance results over the 3 periods was examined. The trends of antibiotic utilization rates and pathogen culture rates in periods 2 and 3 were also analysed.ResultsA total of 114,647 inpatients, including 2242 HAI cases, were analysed. The incidence of HAIs in period 2 was significantly greater than that in period 1 (2.28% vs. 1.48%, χ2 = 61.963, p < 0.001) and period 3 (2.28% vs. 2.05%, χ2 = 4.767, p = 0.029). The incidence of five HAI sites, including respiratory infection, urinary tract infection (UTI), surgical site infection (SSI), bloodstream infection (BSI) and skin and soft tissue infection, was significantly greater in period 2 compared with period 1 (both p < 0.05) but was not significantly different from that in period 3. The incidence of hospital-acquired MDRO infections in period 3 was lower than that in period 2. The identification of MDRO infection cases using the RT-NISS achieved a high level of sensitivity (Se), specificity (Sp), positive predictive value (PPV) and negative predictive value (NPV), especially in period 3 (Se = 100%, Sp = 100%, PPV = 100% and NPV = 100%).ConclusionThe adoption of a RT-NISS to adequately and accurately collect HAI cases is useful to prevent and control HAIs. Furthermore, RT-NISSs improve accuracy in MDRO infection case reporting, which can timely and accurately guide and supervise clinicians in implementing MDRO infection prevention and control measures.

An Automated Surveillance System (SurCeGGID) for the French Sexually Transmitted Infection Clinics: Epidemiological Monitoring Study.

Viral and bacterial sexually transmitted infections (STIs) are public health concerns worldwide, but surveillance systems are not comprehensive enough to design and monitor accurately STI control strategies in most countries. In 2016, 320 STI clinics (CeGIDDs in French) were implemented in France, primarily targeting most exposed populations, although access is free of charge for anybody. This article describes the mandatory surveillance system (SurCeGIDD) based on CeGIDD's individual data aiming to better guide STI prevention. A decree ensured the use of software to manage consultations in CeGIDDs and to transfer surveillance data. A web service was implemented to secure data transfer from CeGIDDs' software to a centralized database. CeGIDDs can also transfer data in CSV format via a secured data-sharing platform. Then, data are automatically checked before integration. Data on sociodemographic variables, sexual exposure, blood exposure, symptoms, STI tests, STI diagnoses, and sexual health services delivery were collected for the previous year (n-1). Preliminary and descriptive analyses of 2017-2018 data transmitted in 2018 and 2019, respectively, were performed using numbers and proportions for qualitative variables. In 2017, 54/320 (16.9%) CeGIDDs transmitted their data. In 2018, this number of participants increased to 143/320 (44.7%) CeGIDDs. The corresponding volume of records increased from 2414 in 2017 to 382,890 in 2018. In 2018, most attendances were hospital based (263,480/382,890, 68.81%). In 2018, attendees were mostly men 227,326/379,921 (59.84%), while 151,963/379,921 (40%) were women 632/379,921 (0.17%) transgenders. The median age was 27 years for men, 23 years for women, and 30 years for transgender. Half of the attendees (81,964/174,932, 46.85%) were heterosexual men, 69,016/174,932 (39.45%) heterosexual women, 20,764/174,932 (11.87%) men who have sex with men, and 3188/174,932 (1.82%) women who have sex with women. A majority of them were born in France (227,698/286,289, 79.53%) and unemployed 115,913/211,707 (54.75%). The positivity rates were 0.37% for 205,348 HIV serologies, 1.31% for 131,551 hepatitis B virus serologies, 7.16% for 161,241 Chlamydia trachomatis PCR, 2.83% for 146,649 gonorrhea PCR, 1.04% for the syphilis combination of treponema and nontreponema serologies, and 5.96% for 13,313 Mycoplasma genitalium PCR. Despite challenges, the effectiveness of the SurCeGIDD surveillance based on routine patients' records was demonstrated. The wide range of information, including socioeconomic determinants, might help to better guide and evaluate the prevention policies and services delivery. However, the growing volumes of information will require adapted tools and algorithms for the data management and analyses.

Estimating COVID-19 cases in Puerto Rico using an automated surveillance system.

Due to concerns regarding limited testing and accuracy of estimation of COVID-19 cases, we created an automated surveillance system called “Puerto Rico Epidemiological Evaluation and Prevention of COVID-19 and Influenza” (PREPCOVI) to evaluate COVID-19 incidence and time trends across Puerto Rico. Automated text message invitations were sent to random phone numbers with Puerto Rican area codes. In addition to reported COVID-19 test results, we used a published model to classify cases from specific symptoms (loss of smell and taste, severe persistent cough, severe fatigue, and skipped meals). Between 18 November 2020, and 24 June 2021, we sent 1,427,241 messages, 26.8% were reached, and 6,975 participants answered questions about the last 30 days. Participants were aged 21–93 years and represented 97.4% of the municipalities. PREPCOVI total COVID-19 cases were higher among women and people aged between 21 and 40 years and in the Arecibo and Bayamón regions. COVID-19 was confirmed, and probable cases decreased over the study period. Confirmed COVID-19 cases ranged from 1.6 to 0.2% monthly, although testing rates only ranged from 30 to 42%. Test positivity decreased from 13.2% in November to 6.4% in March, increased in April (11.1%), and decreased in June (1.5%). PREPCOVI total cases (6.5%) were higher than cases reported by the Puerto Rico Department of Health (5.3%) for similar time periods, but time trends were similar. Automated surveillance systems and symptom-based models are useful in estimating COVID-19 cases and time trends, especially when testing is limited.

Life-history genotype explains variation in migration activity in Atlantic salmon (Salmo salar).

One of the most well-known life-history continuums is the fast-slow axis, where 'fast' individuals mature earlier than 'slow' individuals. 'Fast' individuals are predicted to be more active than 'slow' individuals because high activity is required to maintain a fast life-history strategy. Recent meta-analyses revealed mixed evidence for such integration. Here, we test whether known life-history genotypes differ in activity expression by using Atlantic salmon (Salmo salar) as a model. In salmon, variation in Vgll3, a transcription cofactor, explains approximately 40% of variation in maturation timing. We predicted that the allele related to early maturation (vgll3*E) would be associated with higher activity. We used an automated surveillance system to follow approximately 1900 juveniles including both migrants and non-migrants (i.e. smolt and parr fish, respectively) in semi-natural conditions over 31 days (approx. 580 000 activity measurements). In migrants, but not in non-migrants, vgll3 explained variation in activity according to our prediction in a sex-dependent manner. Specifically, in females the vgll3*E allele was related to increasing activity, whereas in males the vgll3*L allele (later maturation allele) was related to increasing activity. These sex-dependent effects might be a mechanism maintaining within-population genetic life-history variation.

Integration of Biobanking Architecture with Genomics Data: Genomics Integrated Biobanking Ontology (GIBO).

Integration of clinical-pathological information of Biobanks with genomics-epidemiological data/inferences in a structured and consistent manner, mitigating inherent heterogeneities of sites/sources of data/sample collection, processing, and information storage hurdles, is primary to achieving an automated surveillance system. Genomics Integrated Biobanking Ontology (GIBO) presents a solution for preserving the contextual meaning of heterogeneous data, while interlinking different genomics and epidemiological concepts in machine comprehensible format with the biobank framework. GIBO an OWL ontology introduces 84 new classes to integrate genomics data relevant to public health.

Passenger Surveillance Using Deep Learning in Post-COVID-19 Intelligent Transportation System.

Intelligent Transport System should be renovated in many aspects in post-pandemic situation like COVID-19. The passenger-count inside a car will be restricted based on the vehicle capacity and the COVID-19 hot-spot zone. Traffic rules will be impacted to align with a similar contagious outbreak. The on-road ‘Yellow-Vulture’ cameras need to incorporate such surveillance rules to monitor related anomalies for preventing contamination. To maintain safe-distance, an automatic surveillance system will be preferred by the Government very soon. Moreover, facial mask usage during the journey has become an essential habit to stop the spread of the infection. In this article, we have proposed a deep-Learning based framework that employs an augmented image data set to provide proper surveillance in the transport system to maintain the health protocols. Fast and accurate detection of the number of passengers inside a car and their face masks from the traffic inspection camera feed has been demonstrated. We have exploited the advantages of the popular Transfer Learning approach with novel variations of images while performing the training. To the best of our knowledge, this is the first attempt to watch over in-vehicle social-distancing in post-pandemic circumstances through deep-Learning based image analysis. The superiority of the proposed framework has been established over several state-of-the-art techniques using different numerical metrics and visual comparisons along with a support of statistical hypothesis test. Our technique has achieved 98.5% testing accuracy in various adverse conditions. Zero-shot evaluation has been explored for the Real-Time-Medical-Mask-Detection data set Wang et al. (Real-Time-Medical-Mask-Detection, 2020a https://github.com/TheSSJ2612/Real-Time-Medical-Mask-Detection/, Accessed 14 Nov 2020), where we have attained 96.4% accuracy that manifests the generalization of the network.

Computer Vision based Surveillance System

Abstract: Automated surveillance systems are surveillance systems that employ cameras to monitor the area. The observed setting is analyzed using motion detection, crowd behavior, individual behavior, and interactions between people, crowds, and their surroundings. These automated systems are capable of a wide range of tasks, including detection, interpretation, understanding, recording, and alert creation based on the results of the study. Researchers have been able to increase monitoring performance while limiting human errors by changing various features of these systems. This research looks into video surveillance systems and the components that go with them in depth. The designs that were used, as well as the most important studies in these systems, are presented. Existing surveillance systems were compared in terms of to provide a wider picture and a more holistic understanding of the system.

DeepFire: A Novel Dataset and Deep Transfer Learning Benchmark for Forest Fire Detection

Forest fires pose a potential threat to the ecological and environmental systems and natural resources, impacting human lives. However, automated surveillance system for early forest fire detection can mitigate such calamities and protect the environment. Therefore, we propose a UAV-based forest fire fighting system with integrated artificial intelligence (AI) capabilities for continuous forest surveillance and fire detection. The major contributions of the proposed research are fourfold. Firstly, we explain the detailed working mechanism along with the key steps involved in executing the UAV-based forest fire fighting system. Besides, a robust forest fire detection system requires precise and efficient classification of forest fire imagery against no-fire. Moreover, we have curated a novel dataset (DeepFire) containing diversified real-world forest imagery with and without fire to assist future research in this domain. The DeepFire dataset consists of 1900 colored images with 950 each for fire and no-fire classes. Next, we investigate the performance of various supervised machine learning classifiers for the binary classification problem of detecting forest fire. Furthermore, we propose a VGG19-based transfer learning solution to achieve improved prediction accuracy. We assess and compare the performance of several machine learning approaches such as k -nearest neighbors, random forest, naive Bayes, support vector machine, logistic regression, and the proposed approach for accurately identifying fire and no-fire images in the DeepFire dataset. The simulation results demonstrate the efficacy of the proposed approach in terms of accurate classification, where it achieves the mean accuracy of 95% with 95.7% precision and 94.2% recall.

A Low Power Architectural Framework for Automated Surveillance System with Low Bit Rate Transmission

Abstract The changed security scenario of the modern time has necessitated increased and sophisticated vigilance of the countries' borders. The technological challenges involved in accomplishing such feat of automated security system are many and require research at the components-and-algorithms as well as the architectural levels. This paper proposes an architectural framework for automated video surveillance comprising a network of sensors and closed circuit television cameras as well as proposing algorithmic/component research of software and hardware for the core functioning of the framework, such as: communication protocols, object detection, data-integration, object identification, object tracking, video compression, threat identification, and alarm generation. In this paper, we are addressing some general topological and routing features that would be adopted in our system. There are two types of data with regard to data communication – video stream and object detection. The network is broken down into several disjoint, almost equal zones. A zone have one or more one cluster. A zone manager is chosen among the cluster heads depending on their relative residual energies. There are several levels of control that could be implemented with this arrangement with localized decision made, to get distributed effect at all levels. A cell tracks each target in its zone. If the target moves out of the range of a cell, the cell manager will send the target description to estimated next cell. The next cell starts tracking the target. If the estimated cell is wrongly chosen, corrections will be made by the cluster heads to get the new target-tracking. We also propose bitrate reduction algorithms to accommodate the limited bandwidth. One of the main feature of this paper is introducing a Low-Power Low-Bit rate video compression algorithm to accommodate the low power requirements at sensor nodes, and the low bit rate requirement for the communication protocol. We proposed two algorithms the ALBR and LPHSME. ALBR is addressing low bit rate required for sensors network with limited bandwidth which achieves a reduction in Average number of bits per Iframe by approximately 60% in case of low motion video sequences and 53% in case of fast motion video sequences . LPHSME addresses low power requirements of multi sensor network that has limited power batteries. The performance of the proposed LPHSME algorithm versus full search and three step search indicates a reduction in motion estimation time by approximately 89% in case of low motion video sequences (e.g., Claire ) and 84% in case of fast motion video sequences. The reduced complexity of LPHSME results in low power requirements.

Convolutional Neural Networks Hyperparameter Tunning for Classifying Firearms on Images

ABSTRACT In recent years, the rates of firearm-related violent acts worldwide have risen significantly. It is a severe social problem that compromises the safety of every individual to some extent. This situation has motivated researchers to find new ways to improve the current state-of-the-art solutions, such as automatic surveillance systems, to detect and classify the presence of firearms within a specific scene. These systems reduce the drawbacks of using direct human supervision. Among the available solutions for the classification task, the performance of Deep Learning models stands out, especially those based on Convolutional Neural Networks. Since they start learning directly from raw data (e.g., images), their learning process can be improved even further by using Transfer Learning techniques. However, the classification accuracy depends significantly on choosing the optimum set of values for the different hyperparameters composing them. Thus, this paper analyses the improvement in the performance of an image-based handgun classification algorithm when tuning its hypermeters values instead of using its default values. In this work, we evaluated the performance variation using two benchmarks Convolutional Neural Networks architectures: AlexNet and Inception V3. We obtained a maximum accuracy of 94.11% when using the Inception V3 network and transfer learning. We employed Nadam as the optimizer and a learning rate equal to 0.0001, a batch size equal 256, and a total of 13 epochs. Experimental results suggest an essential relationship between the performance of the classification model and the data set, the specific combinations of values for the selected optimizer, the batch size, and the learning rate. The obtained improvement in the accuracy was up to 10.33% after the tuning process.