Real-time Object Detection System Research Articles

Improving video surveillance systems in banks using deep learning techniques

In the contemporary world, security and safety are significant concerns for any country that wants to succeed in tourism, attracting investors, and economics. Manually, guards monitoring 24/7 for robberies or crimes becomes an exhaustive task, and real-time response is essential and helpful for preventing armed robberies at banks, casinos, houses, and ATMs. This paper presents a study based on real-time object detection systems for weapons auto-detection in video surveillance systems. We propose an early weapon detection framework using state-of-the-art, real-time object detection systems such as YOLO and SSD (Single Shot Multi-Box Detector). In addition, we considered closely reducing the number of false alarms in order to employ the model in real-life applications. The model is suitable for indoor surveillance cameras in banks, supermarkets, malls, gas stations, and so forth. The model can be employed as a precautionary system to prevent robberies by implying the model in outdoor surveillance cameras.

Real-time Face and Object Detection with Age and Gender Prediction for Video Surveillance Applications

Abstract: The development of advanced computer vision techniques has made it possible to perform real-time face and object detection for video surveillance applications. Video surveillance is an essential tool for monitoring public areas, buildings, and other locations for security and safety purposes. However, analyzing the vast amount of data generated by video surveillance systems can be challenging. Real-time face detection and object detection systems provide an effective solution to this problem, enabling the identification and tracking of people and objects of interest. In recent years, there has been a growing interest in incorporating age prediction and gender prediction into video surveillance systems. The ability to predict the age and gender of individuals can provide valuable insights to enhance the effectiveness of video surveillance applications. For example, age and gender prediction can be used to detect and prevent potential crimes. This research paper presents a study of a real-time face and object detection system with age and gender prediction for video surveillance applications. The proposed system utilizes deep learning and convolutional neural network techniques to achieve high accuracy in face and object detection, as well as age and gender prediction. The experimental results demonstrate the effectiveness of the proposed system in accurately detecting and tracking faces and objects while predicting their age and gender in real time. The proposed system has potential applications in various fields

INVESTIGATING THE PERFORMANCE INDICES OF YOLO MODELS IMPLEMENTED ON A DPU: A COMPARATIVE ANALYSIS

A comprehensive analysis of the performance indices achieved by implementing various “You Only Look Once” dataset (YOLO) models on a dedicated Deep Learning Processing Unit (DPU). YOLO models are renowned for their real-time object detection capabilities, making them a good choice for a range of applications including autonomous vehicles, surveillance systems, and robotics. In this study, the YOLO models are deployed onto a specialized hardware accelerator, specifically a DPU, to assess their inference speed, accuracy, and power efficiency. By conducting an in-depth comparative evaluation of multiple YOLO variants, including YOLOv3, YOLOv4, YOLOv5, YOLOv6 insights into how each model interacts with the DPU architecture are revealed. The experiments involve benchmarking these models across diverse datasets and varying hardware configurations. The results not only highlight the advantages and limitations of employing DPUs for YOLO-based applications but also help in choosing the most suitable model-DPU combination based on specific performance requirements. This study contributes to the optimization of real-time object detection systems and assists practitioners in making informed decisions regarding the model and hardware selection.

Design and implementation of real-time object detection system based on single-shoot detector and OpenCV.

Computer vision (CV) and human-computer interaction (HCI) are essential in many technological fields. Researchers in CV are particularly interested in real-time object detection techniques, which have a wide range of applications, including inspection systems. In this study, we design and implement real-time object detection and recognition systems using the single-shoot detector (SSD) algorithm and deep learning techniques with pre-trained models. The system can detect static and moving objects in real-time and recognize the object's class. The primary goals of this research were to investigate and develop a real-time object detection system that employs deep learning and neural systems for real-time object detection and recognition. In addition, we evaluated the free available, pre-trained models with the SSD algorithm on various types of datasets to determine which models have high accuracy and speed when detecting an object. Moreover, the system is required to be operational on reasonable equipment. We tried and evaluated several deep learning structures and techniques during the coding procedure and developed and proposed a highly accurate and efficient object detection system. This system utilizes freely available datasets such as MS Common Objects in Context (COCO), PASCAL VOC, and Kitti. We evaluated our system's accuracy using various metrics such as precision and recall. The proposed system achieved a high accuracy of 97% while detecting and recognizing real-time objects.

An Intelligent Real-Time Object Detection System on Drones

Drones have been widely used in everyday life and they can help deal with various tasks, including photography, searching, and surveillance. Nonetheless, it is difficult for drones to perform customized online real-time object detection. In this study, we propose an intelligent real-time object detection system for drones. It is composed of an FPGA and a drone. A neural-network (NN) engine is designed on the FPGA for NN model acceleration. The FPGA receives activation data from an NN model, which are assembled into the data stream. Multiple fetch and jump pointers catch required activation values from the data stream, which are then filtered and sent to each thread independently. To accelerate processing speed, multiple processing elements (PEs) deal with tasks in parallel by using multiple weights and threads. The image data are transferred from the drone host to the FPGA, which are tackled with high speed by the NN engine. The NN engine results are returned to the host, which is used to adjust the flying route accordingly. Experimental results reveal that our proposed FPGA design well utilizes FPGA computing resources with 81.56% DSP and 72.80% LUT utilization rates, respectively. By using the Yolov3-tiny model for fast object detection, our system can detect objects at the speed of 8 frames per second and achieves a much lower power consumption compared to state-of-the-art methods. More importantly, the intelligent object detection techniques provide more pixels for the target of interest and they can increase the detection confidence score from 0.74 to 0.90 and from 0.70 to 0.84 for persons and cars, respectively.

DRONE DETECTION USING DEEP LEARNING

Drones have widespread application in real life and the industry is expanding rapidly. As they are growing increasingly it is more accessible to the public at cheaper prices. They are used for espionage and can be converted into gruesome weapons. Hence it is very important to monitor and detect unauthorized drones entering into the restricted regions in order to maintain peace and prevent chaos. The Technology stack implied here is You Only Look Once (YOLO v5) which is a real time object detection system. In recent times Yolo is a profound algorithm used for real time object or image detection . The Yolo trained model is trained with pictures of drones and birds so that the trained model can differentiate between and prevent false prediction of drones . Everytime a drone is detected in the camera an alert message is sent to the higher officials so that the drone could be eliminated. This algorithm comprises 3 techniques namely: Residual blocks, Bounding box regression and Intersection over Union. The camera used here is 360 degree so that maximum area of visibility is obtained. KEYWORDS: Drones, real time object detection , YOLO v4.

Design of Real-Time Object Detection in Mobile Robot for Volcano Monitoring Application

Indonesia is one of the countries located at the ring of fire which should be monitored to predict the eruption earlier and set the risk zones around with no human involvement especially while eruption taking place. Therefore, in this research, it is used a 4 wheeled mobile robot called PRAWIRA for this purpose. The robot should have the ability to avoid the obstacles in front of it in this area. It has been designed a real-time object detection system for volcano monitoring application using deep learning from the YOLOv5s model for 4 objects (trees, persons, stones, and stairs). It was used 484 images for the dataset after the pre-train process was conducted with several steps: object identification; dataset downloading (Google Chrome Extension and Open Images v6); image labeling (LabeImg); augmentation process (flip, blur, and rotation); and data training for varies epochs and batches by Jupyter Notebook GPU. The preliminary result for this research was presented in the mean average precision (mAP) of YOLOv5s (the smallest version). The first variation (batch = 16, epochs = 100) resulted in mAP_0.5 = 17.9% and mAP_0.5:0.95 = 7.27% with 0.262 hours of training time. The second (batch = 16, epochs = 500) resulted in mAP_0.5 = 25.7% and mAP_0.5:0.95 = 12.3% with 1.296 hours of training time, while the third (batch = 80, epochs = 100) resulted in mAP_0.5 = 17.7% and mAP_0.5:0.95 = 5.63% with 0.232 hours of training time. Furthermore, the last variation (batch = 80, epochs = 500) resulted in mAP_0.5 = 19.5% and mAP_0.5:0.95 = 8.92% with 1.085 hours of training time. Therefore, the second variation is the best result for the model with 14.8 MB of size. Moreover, interfaces for the best model were displayed to show the result of the training.

FaceMask: A New Image Dataset for the Automated Identification of People Wearing Masks in the Wild.

The rapid spread of the COVID-19 pandemic, in early 2020, has radically changed the lives of people. In our daily routine, the use of a face (surgical) mask is necessary, especially in public places, to prevent the spread of this disease. Furthermore, in crowded indoor areas, the automated recognition of people wearing a mask is a requisite for the assurance of public health. In this direction, image processing techniques, in combination with deep learning, provide effective ways to deal with this problem. However, it is a common phenomenon that well-established datasets containing images of people wearing masks are not publicly available. To overcome this obstacle and to assist the research progress in this field, we present a publicly available annotated image database containing images of people with and without a mask on their faces, in different environments and situations. Moreover, we tested the performance of deep learning detectors in images and videos on this dataset. The training and the evaluation were performed on different versions of the YOLO network using Darknet, which is a state-of-the-art real-time object detection system. Finally, different experiments and evaluations were carried out for each version of YOLO, and the results for each detector are presented.

Deep Neural Networks for Gun Detection in Public Surveillance

The conventional surveillance and control system of Closed-Circuit Television (CCTV) cameras require human resource supervision. Almost all the criminal activities take place using weapons mostly handheld gun, revolver, or pistol. Automatic gun detection is a vital requirement now-a-days. The use of real-time object detection system for the improvement of surveillance is a promising application of Convolutional Neural Networks (CNN). We are concerned about the real-time detection of weapons for the surveillance cameras, so we focused on the implementation and comparison of faster approaches such as Region (R-CNN) and Region Fully Convolutional Networks (R-FCN) with feature extractor Visual Geometry Group (VGG) and ResNet respectively. Training and testing are done on database that consists of local environment images. These images are taken with different type and high- resolution cameras that minimize the idealism. Some metrics also defined to reduce the false positives which are specific to the solution of problem. This research also contributes to the constitution of a hybrid CNN model of both faster-based R-CNN and R-FCN. Both hybrid and existing models experimented to reduce false positive in weapon detection. Result represented in graph with calculation during and after training with confusion matrix and hybrid model results better than other models.

A Comparative Study of Real-time Object Detection Systems for Navigation of the Visually Impaired

The visually impaired face a plethora of problems. The primary problem they face is navigating from one place to another. The detection of obstacles in the user's proximity is another challenge that needs to be addressed. This paper provides a comparative study of various real-time image recognition and object detection methods that might help develop effective navigation systems for the visually impaired.

Detecting pelvic fracture on 3D-CT using deep convolutional neural networks with multi-orientated slab images

Pelvic fracture is one of the leading causes of death in the elderly, carrying a high risk of death within 1 year of fracture. This study proposes an automated method to detect pelvic fractures on 3-dimensional computed tomography (3D-CT). Deep convolutional neural networks (DCNNs) have been used for lesion detection on 2D and 3D medical images. However, training a DCNN directly using 3D images is complicated, computationally costly, and requires large amounts of training data. We propose a method that evaluates multiple, 2D, real-time object detection systems (YOLOv3 models) in parallel, in which each YOLOv3 model is trained using differently orientated 2D slab images reconstructed from 3D-CT. We assume that an appropriate reconstruction orientation would exist to optimally characterize image features of bone fractures on 3D-CT. Multiple YOLOv3 models in parallel detect 2D fracture candidates in different orientations simultaneously. The 3D fracture region is then obtained by integrating the 2D fracture candidates. The proposed method was validated in 93 subjects with bone fractures. Area under the curve (AUC) was 0.824, with 0.805 recall and 0.907 precision. The AUC with a single orientation was 0.652. This method was then applied to 112 subjects without bone fractures to evaluate over-detection. The proposed method successfully detected no bone fractures in all except 4 non-fracture subjects (96.4%).

Object detection YOLO-v4 based smart car system with an integrated Virtual Assistant

Car accidents are a serious social problem that often results in both life loss and financial loss. Most car accidents are caused by a lack of safe distance between cars and also the awareness of the driver. To relieve this issue, in this paper we propose a real-time object detection and safety system. The proposed system consists of two units: a real-time object detection unit and a safety alarm unit. The system is supposed to apply in a normal driving scenario. As for the safety alarm module, it consists of three states: to detect the object; to calculate the safety factor; to determine the driving conditions. To justify the proposed system, a real experiment is conducted. The results show that the system can appropriately signify driving states: safe, dangerous, and warning. By the given experimental results, it implies that the system is feasible and applicable in real-time applications. The objective of this project is to detect objects in real-time and pass information to the user based on the data taken by object detection. The information can be provided to warn the user or to just inform the user based on the result of the sensors. The YOLO algorithm helps to design the output of the sensor based on the necessity of the user. The google assistant allows the user to provide voice commands. The car is integrated with all the above-mentioned.

Currency Detection for Visually Impaired Iraqi Banknote as a Study Case

It is relatively simple for a normal human to interpret and understand every banknote, but one of the major problems for visually impaired people are money recognition, especially for paper currency.
 Since money plays such an important role in our everyday lives and is required for every business transaction, real-time detection and recognition of banknotes become a necessity for blind or visually impaired people For that purpose, we propose a real-time object detection system to help visually impaired people in their daily business transactions.
 Dataset Images of the Iraqi banknote category are collected in different conditions initially and then, these images are augmented with different geometric transformations, to make the system strong. These augmented images are then annotated manually using the "LabelImg" program, from which training sets and validation image sets are prepared.
 We will use YOLOv3 real-time Object Detection algorithm trained on custom Iraqi banknote dataset for detection and recognition of banknotes. Then the label of the banknotes is identified and then converted into audio by using Google Text to Speech (gTTS), which will be the expected output.
 The performance of the trained model is evaluated on a test dataset and real-time live video. The test results demonstrate that the proposed method can detect and recognize Iraqi paper money with high mAP reaches 97.405% and a short time.

Lightweight Feature Enhancement Network for Single-Shot Object Detection.

At present, the one-stage detector based on the lightweight model can achieve real-time speed, but the detection performance is challenging. To enhance the discriminability and robustness of the model extraction features and improve the detector’s detection performance for small objects, we propose two modules in this work. First, we propose a receptive field enhancement method, referred to as adaptive receptive field fusion (ARFF). It enhances the model’s feature representation ability by adaptively learning the fusion weights of different receptive field branches in the receptive field module. Then, we propose an enhanced up-sampling (EU) module to reduce the information loss caused by up-sampling on the feature map. Finally, we assemble ARFF and EU modules on top of YOLO v3 to build a real-time, high-precision and lightweight object detection system referred to as the ARFF-EU network. We achieve a state-of-the-art speed and accuracy trade-off on both the Pascal VOC and MS COCO data sets, reporting 83.6% AP at 37.5 FPS and 42.5% AP at 33.7 FPS, respectively. The experimental results show that our proposed ARFF and EU modules improve the detection performance of the ARFF-EU network and achieve the development of advanced, very deep detectors while maintaining real-time speed.

Real Time Object Detection Using SSD For Bank Security

Object detection in real time applications are more challenging and thus faces more difficulties in designing object detection applications. Object detection is a method of identifying predefined objects from images, videos or from other input sources. It consist of different steps for identifying the object from its surroundings. Earlier detection system uses different techniques for objection detection with computationally demanding networks that requires high GPU and needs more power. Considering the current circumstances in the society, it is better to design real time object detection system for security applications. The newly evolved technique such as Tensorflow(TF) along with OpenCV can be used used for object detection. The CCTV footages can be used as input and the monitoring of it can be done inorder to find any security violations. The proposed real time object detection system for bank security mainly aims in providing security to bank vaults. This system can also be used for other security applications where human entry is restricted. The proposed model can be used for different security related applications effectively.

Saliency-based YOLO for single target detection

At present, You only look once (YOLO) is the fastest real-time object detection system based on a unified deep neural network. During training, YOLO divides the input image to $$S \times S $$ gird cells and the only one grid cell that contains the center of an object, takes charge of detecting that object. It is not sure that the cell corresponding to the center of the object is the best choice to detect the object. In this paper, inspired by the visual saliency mechanism we introduce the saliency map to YOLO to develop YOLO3-SM method, where saliency map selects the grid cell containing the most salient part of the object to detect the object. The experimental results on two data sets show that the prediction box of YOLO3-SM obtains the lager IOU value, which demonstrates that compared with YOLO3 , the YOLO3-SM selects the cell that is more suitable to detect the object . In addition, YOLO3-SM gets the highest mAP that the other three state-of-the-art object detection methods on the two data sets, which shows that introducing the saliency map to YOLO can improve the detection performance.

Detection of COVID-19 cases through X-ray images using hybrid deep neural network

PurposeThe latest 2019 coronavirus (COVID-2019), which first appeared in December 2019 in Wuhan's city in China, rapidly spread around the world and became a pandemic. It has had a devastating impact on daily lives, the public's health and the global economy. The positive cases must be identified as soon as possible to avoid further dissemination of this disease and swift care of patients affected. The need for supportive diagnostic instruments increased, as no specific automated toolkits are available. The latest results from radiology imaging techniques indicate that these photos provide valuable details on the virus COVID-19. User advanced artificial intelligence (AI) technologies and radiological imagery can help diagnose this condition accurately and help resolve the lack of specialist doctors in isolated areas. In this research, a new paradigm for automatic detection of COVID-19 with bare chest X-ray images is displayed. Images are presented. The proposed model DarkCovidNet is designed to provide correct binary classification diagnostics (COVID vs no detection) and multi-class (COVID vs no results vs pneumonia) classification. The implemented model computed the average precision for the binary and multi-class classification of 98.46% and 91.352%, respectively, and an average accuracy of 98.97% and 87.868%. The DarkNet model was used in this research as a classifier for a real-time object detection method only once. A total of 17 convolutionary layers and different filters on each layer have been implemented. This platform can be used by the radiologists to verify their initial application screening and can also be used for screening patients through the cloud.Design/methodology/approachThis study also uses the CNN-based model named Darknet-19 model, and this model will act as a platform for the real-time object detection system. The architecture of this system is designed in such a way that they can be able to detect real-time objects. This study has developed the DarkCovidNet model based on Darknet architecture with few layers and filters. So before discussing the DarkCovidNet model, look at the concept of Darknet architecture with their functionality. Typically, the DarkNet architecture consists of 5 pool layers though the max pool and 19 convolution layers. Assume as a convolution layer, and as a pooling layer.FindingsThe work discussed in this paper is used to diagnose the various radiology images and to develop a model that can accurately predict or classify the disease. The data set used in this work is the images bases on COVID-19 and non-COVID-19 taken from the various sources. The deep learning model named DarkCovidNet is applied to the data set, and these have shown signification performance in the case of binary classification and multi-class classification. During the multi-class classification, the model has shown an average accuracy 98.97% for the detection of COVID-19, whereas in a multi-class classification model has achieved an average accuracy of 87.868% during the classification of COVID-19, no detection and Pneumonia.Research limitations/implicationsOne of the significant limitations of this work is that a limited number of chest X-ray images were used. It is observed that patients related to COVID-19 are increasing rapidly. In the future, the model on the larger data set which can be generated from the local hospitals will be implemented, and how the model is performing on the same will be checked.Originality/valueDeep learning technology has made significant changes in the field of AI by generating good results, especially in pattern recognition. A conventional CNN structure includes a convolution layer that extracts characteristics from the input using the filters it applies, a pooling layer that reduces calculation efficiency and the neural network's completely connected layer. A CNN model is created by integrating one or more of these layers, and its internal parameters are modified to accomplish a specific mission, such as classification or object recognition. A typical CNN structure has a convolution layer that extracts features from the input with the filters it applies, a pooling layer to reduce the size for computational performance and a fully connected layer, which is a neural network. A CNN model is created by combining one or more such layers, and its internal parameters are adjusted to accomplish a particular task, such as classification or object recognition.

Real-time object detection and face recognition system to assist the visually impaired

The application system presented here is for use in an Android based mobile phone by the visually impaired for detecting objects in their vicinity, which will help them move around safely, without crashing into objects. The detection of objects is done from a real time video taken from the mobile phone camera. People and objects are detected from this video using OpenCV, YOLO and FaceNet. If a human is detected, the system identifies the person. His identity will then be converted into audio and presented to the user. Similarly, objects detected in the vicinity will be presented in audio format to the user.

Towards a physical-world adversarial patch for blinding object detection models

As one of the core components of the computer vision, the object detection model plays a vital role in various security-sensitive systems. However, it has been proved that the object detection model is vulnerable to the adversarial attack. In this paper, we propose a novel adversarial patch attack against object detection models. Our attack can make the object of a specific class invisible to object detection models. We design the detection score to measure the detection model’s output and generate the adversarial patch by minimizing the detection score. We successfully suppress the model’s inference and fool several state-of-the-art object detection models. We triumphantly achieve a minimum recall of 11.02% and a maximum fooling rate of 81.00% and demonstrates the high transferability of adversarial patch between different architecture and datasets. Finally, we successfully fool a real-time object detection system in the physical world, demonstrating the feasibility of transferring the digital adversarial patch to the physical world. Our work illustrates the vulnerability of the object detection model against the adversarial patch attack in both the digital and physical world.

A RGB-D Based Real-Time Multiple Object Detection and Ranging System for Autonomous Driving

Real-time object detection and ranging of multiple objects on the road are the essential tasks in the field of autonomous driving. In this paper, we introduce a system for simultaneous detection and ranging of vehicles, people, non-motor vehicles and lanes based on RGB-D images. Among them, the detection of vehicles, people and non-motor vehicles belongs to general detection task and the lane detection belongs to segmentation task. In order to improve the accuracy and speed, we use two networks to complete these two tasks. We propose a real-time synchronization method with multi-GPU, which achieves separate training and simultaneous detection of lane detectior and vehicle, people and non-motor vehicle detector. We also propose a center-selective ranging module based on binocular ranging technology to distance the detected object. The system reaches nearly 15 FPS with four 1080Ti GPUs. We construct datasets about these problems including daytime and night in which the system achieves high accuracy. A real-time test of the system on the streets of Tianjin, China has been conducted by us, it has proved that the system can be applied to actual driving.