YOLO V3 Research Articles

Green-Tech CAV: Next Generation Computing for Traffic Sign and Obstacle Detection in Connected and Autonomous Vehicles

In recent years, advanced technologies are used in automobiles, which make it simpler to drive on roads. However, driving is not trouble-free considering the unexpected hindrances and difficulty in recognizing traffic signs while driving at high speeds. Due to the advancement in technologies such as image recognition, several automated vehicles and techniques are developed to assist the drivers while driving. In this paper, an integrated intelligent system is designed using YOLO V3 neural architecture for recognizing the obstacles and visualizing the traffic sign through the input image from the camera of the moving vehicle. A Vehicular Adhoc Network (VANET) is being used with 5G connectivity to share the information to the other passing vehicles present in the vicinity. Apart from this, a notification is sent to the nearby centre for clearing any obstacle if found on roads during transportation. Real-time images of every possible obstacle such as a crack on roads along with traffic signs are trained using machine learning algorithms, thereby making the system more efficient and accurate. These techniques are developed in order to provide an pollution free environment and reduce accidents in roads with the concepts of green computing. The proposed automated system performs exemplary in recognizing the obstacle and traffic sign, making it easy for the drivers to drive more freely and reducing the pollution thereby producing a eco-friendly environment. According to the algorithms proposed in this paper, the results show that the algorithm used produces accurate results compared to the existing techniques.

An Efficient Scheme to Obtain Background Image in Video for YOLO-based Static Object Recognition

Detecting backgrounds in videos is an important technology that can be used for many applications such as management of major facilities and military surveillance depending on the purpose. It is difficult to accurately find and identify important objects in the background if there are obstacles such as pedestrian or car in the video. In order to overcome this problem, the following method is used to detect the background. First, a pixel area histogram is generated to determine the amount of change in pixel units of an image over time. Based on the histogram, we propose an algorithm that estimates the background by selecting the case with the smallest rate of change. In addition, in order to strongly respond to changes in the surrounding environment, even when a change in brightness occurs, this is solved through frame overlap. Finally, the desired object is identified by applying YOLO v3 as a model for object detection in the obtained background. Through the above process, this study proposes a method for effectively identifying static objects in the background by precisely estimated background of the video. Experimental results show that the non-detection and false detection rate for the background object is enhanced by 60.2% and 11.2%, respectively, in comparison with when the proposed method was not applied.

Quantitative Measurement of Spinal Cerebrospinal Fluid by Cascade Artificial Intelligence Models in Patients with Spontaneous Intracranial Hypotension.

Cerebrospinal fluid (CSF) hypovolemia is the core of spontaneous intracranial hypotension (SIH). More than 1000 magnetic resonance myelography (MRM) images are required to evaluate each subject. An effective spinal CSF quantification method is needed. In this study, we proposed a cascade artificial intelligence (AI) model to automatically segment spinal CSF. From January 2014 to December 2019, patients with SIH and 12 healthy volunteers (HVs) were recruited. We evaluated the performance of AI models which combined object detection (YOLO v3) and semantic segmentation (U-net or U-net++). The network of performance was evaluated using intersection over union (IoU). The best AI model was used to quantify spinal CSF in patients. We obtained 25,603 slices of MRM images from 13 patients and 12 HVs. We divided the images into training, validation, and test datasets with a ratio of 4:1:5. The IoU of Cascade YOLO v3 plus U-net++ (0.9374) was the highest. Applying YOLO v3 plus U-net++ to another 13 SIH patients showed a significant decrease in the volume of spinal CSF measured (59.32 ± 10.94 mL) at disease onset compared to during their recovery stage (70.61 ± 15.31 mL). The cascade AI model provided a satisfactory performance with regard to the fully automatic segmentation of spinal CSF from MRM images. The spinal CSF volume obtained through its measurements could reflect a patient’s clinical status.

A Lightweight Neural Network-Based Method for Detecting Estrus Behavior in Ewes

We propose a lightweight neural network-based method to detect the estrus behavior of ewes. Our suggested method is mainly proposed to solve the problem of not being able to detect ewe estrus behavior in a timely and accurate manner in large-scale meat sheep farms. The three main steps of our proposed methodology include constructing the dataset, improving the network structure, and detecting the ewe estrus behavior based on the lightweight network. First, the dataset was constructed by capturing images from videos with estrus crawling behavior, and the data enhancement was performed to improve the generalization ability of the model at first. Second, the original Darknet-53 was replaced with the EfficientNet-B0 for feature extraction in YOLO V3 neural network to make the model lightweight and the deployment easier, thus shortening the detection time. In order to further obtain a higher accuracy of detecting the ewe estrus behavior, we joined the feature layers to the SENet attention module. Finally, the comparative results demonstrated that the proposed method had higher detection accuracy and FPS, as well as a smaller model size than the YOLO V3. The precision of the proposed scheme was 99.44%, recall was 95.54%, F1 value was 97%, AP was 99.78%, FPS was 48.39 f/s, and Model Size was 40.6 MB. This study thus provides an accurate, efficient, and lightweight detection method for the ewe estrus behavior in large-scale mutton sheep breeding.

Combination of Deep Cross-Stage Partial Network and Spatial Pyramid Pooling for Automatic Hand Detection

The human hand is involved in many computer vision tasks, such as hand posture estimation, hand movement identification, human activity analysis, and other similar tasks, in which hand detection is an important preprocessing step. It is still difficult to correctly recognize some hands in a cluttered environment because of the complex display variations of agile human hands and the fact that they have a wide range of motion. In this study, we provide a brief assessment of CNN-based object identification algorithms, specifically Densenet Yolo V2, Densenet Yolo V2 CSP, Densenet Yolo V2 CSP SPP, Resnet 50 Yolo V2, Resnet 50 CSP, Resnet 50 CSP SPP, Yolo V4 SPP, Yolo V4 CSP SPP, and Yolo V5. The advantages of CSP and SPP are thoroughly examined and described in detail in each algorithm. We show in our experiments that Yolo V4 CSP SPP provides the best level of precision available. The experimental results show that the CSP and SPP layers help improve the accuracy of CNN model testing performance. Our model leverages the advantages of CSP and SPP. Our proposed method Yolo V4 CSP SPP outperformed previous research results by an average of 8.88%, with an improvement from 87.6% to 96.48%.

VLC localization: deep learning models by Kalman filter algorithm combined with RSS

In this paper, a new framework is presented for indoor visible light communication (VLC) system, based on Yolo v3, EfficientNetB3, and DenseNet121 deep learning (DL) models, as well as an optimization strategy. The proposed framework consists of two steps: data collecting and DL model training. To start, data is acquired using MATLAB and Kalman Filtering (KF) with averaging approaches. Second, the received signal strength (RSS) is employed as the DL models input, with the Cartesian coordinates as the DL models output. The averaging RSS approach combined with KF algorithm are used in the suggested framework. This work introduces the impacts of Non-Line-of-Sight (NLoS) for initial reflection and Line-of-Sight (LoS) based on the three mentioned DL models. Furthermore, we used Bayesian optimization and automatic hyper-parameter (HP) optimization to increase system efficiency and to reduce positioning error in DL models. The obtained results show that the models outperform existing the HP-RSS-KF-LoS-DL models in terms of localization error when compared to traditional RSS signal-based localization techniques. Many performance indicators are considered to evaluate the proposed framework resiliency, including accuracy (ACC), area under the curve (AUC), sensitivity (Se), and precision (Pr), as well as F1-score, root mean square error (RMSE), training, and testing time. The DL models are generated and trained using Python software on a Kaggle Notebook GPU cloud (2 CPU cores and 13 GB RAM). The achieved results are: 99.99% ACC, 99.98% AUC, 98.88% Se, 98.98% Pr, 99.97% F1-score, 0.112 cm RMSE, and 0.29 s testing time. The proposed system could be easily deployed for autonomous applications, based on the analysis of the experimental data. Several applications can be used depending on enhancing the localization of VLC system in military systems, underwater systems, and indoor systems like hospitals, hotels, libraries and malls.

Enhancing Geometric Factors in Model Learning and Inference for Object Detection and Instance Segmentation.

Deep learning-based object detection and instance segmentation have achieved unprecedented progress. In this article, we propose complete-IoU (CIoU) loss and Cluster-NMS for enhancing geometric factors in both bounding-box regression and nonmaximum suppression (NMS), leading to notable gains of average precision (AP) and average recall (AR), without the sacrifice of inference efficiency. In particular, we consider three geometric factors, that is: 1) overlap area; 2) normalized central-point distance; and 3) aspect ratio, which are crucial for measuring bounding-box regression in object detection and instance segmentation. The three geometric factors are then incorporated into CIoU loss for better distinguishing difficult regression cases. The training of deep models using CIoU loss results in consistent AP and AR improvements in comparison to widely adopted ln -norm loss and IoU-based loss. Furthermore, we propose Cluster-NMS, where NMS during inference is done by implicitly clustering detected boxes and usually requires fewer iterations. Cluster-NMS is very efficient due to its pure GPU implementation, and geometric factors can be incorporated to improve both AP and AR. In the experiments, CIoU loss and Cluster-NMS have been applied to state-of-the-art instance segmentation (e.g., YOLACT and BlendMask-RT), and object detection (e.g., YOLO v3, SSD, and Faster R-CNN) models. Taking YOLACT on MS COCO as an example, our method achieves performance gains as +1.7 AP and +6.2 AR100 for object detection, and +1.1 AP and +3.5 AR100 for instance segmentation, with 27.1 FPS on one NVIDIA GTX 1080Ti GPU. All the source code and trained models are available at https://github.com/Zzh-tju/CIoU.

Diagnostic accuracy of deep learning for evaluation of C-spine injury from lateral neck radiographs

BackgroundTraumatic spinal cord injury (TSI) is a leading cause of morbidity and mortality worldwide, with the cervical spine being the most affected. Delayed diagnosis carries a risk of morbidity and mortality. However, cervical spine CT scans are time-consuming, costly, and not always available in general care. In this study, deep learning was used to assess and improve the detection of cervical spine injuries on lateral radiographs, the most widely used screening method to help physicians triage patients quickly and avoid unnecessary CT scans. Materials and methodsLateral neck or lateral cervical spine radiographs were obtained for patients who underwent CT scan of cervical spine. Ground truth was determined based on CT reports. CiRA CORE, a codeless deep learning program, was used as a training and testing platform. YOLO network models, including V2, V3, and V4, were trained to detect cervical spine injury. The diagnostic accuracy, sensitivity, and specificity of the model were calculated. ResultsA total of 229 radiographs (129 negative and 100 positive) were selected for inclusion in our study from a list of 625 patients with cervical spine CT scans, 181 (28.9%) of whom had cervical spine injury. The YOLO V4 model performed better than the V2 or V3 (AUC = 0.743), with sensitivity, specificity, and accuracy of 80%, 72% and 75% respectively. ConclusionDeep learning can improve the accuracy of lateral c-spine or neck radiographs. We anticipate that this will assist clinicians in quickly triaging patients and help to minimize the number of unnecessary CT scans.

Social Distancing Detection in Real Time using Deep Learning and IoT

Abstract: Social Distancing is the practice of maintaining a minimum amount of physical distance from your surrounding people or avoiding direct contact with people or objects in public places so as to minimize the exposure or transmission of diseases like COVID-19. This can also be coupled with wearing of face masks and regular sanitization of hands. The government has made social distancing norms compulsory and to be followed, and penalties would be imposed if violated. But there are a lot of examples of social distancing violations in public places due to negligence of people. There are approaches using bluetooth and mobile phones but it requires an app to be installed. This study provides a method for determining whether the social distancing rule has been broken by combining machine learning models and object detecting techniques. Appropriate actions would be taken in case of any violations detected. In this paper, a comparison is done on all the object detection techniques like Faster R-CNN, YOLO v3, SSD Mobile Net etc. based on accuracy, mean average precision, computational time and ease of integration. Based on this, SSD Mobile Net performed well on accuracy and had a faster computational time. This model can also be deployed on end devices like Raspberry Pi.

Real Time Object Detection Using Deep Learning

Abstract: Visually impaired people have difficulty moving safely and independently, which interferes with normal indoor and outdoor work and social activities. Similarly, they have a hard time identifying the basics of the environment. This paper presents a model for detecting the brightness and key colors of real-time images using the RGB method with an external camera and identifying basic objects and face recognition from human datasets.[2]. Object detection is a department of pc imaginative and prescient that appears for times of lexical entities in photographs and videos. The gadget makes use of the ESP-32 Cam's digital digicam to continuously seize severa frames, which can be sooner or later converted to audio segments. In this project, we use the You Only Look Once V3 (YOLO v3)algorithm, which runs thru a version of a really complex Convolutional Neural Network structure with OpenCV. Then with the aid of using the usage of Google Text to Speech, we convert the photo to textual content and afterwards textual content - to - speech for the visually impaired individual. Thus, the Visually Impaired individual receives the place of the gadgets withinside the digital digicam's view through audio. Distance calculation is aided with the aid of using an ultrasonic sensor. By the usage of The amassed consequences show that the proposed prototype is a hit in presenting visually impaired customers with the cappotential to realise surprising settings the usage of a user-pleasant machine that integrates this unique item detection Model

Forest Fire Detection using Yolov3

Fire detection systems play a key role in industries, shops, malls, etc. in detecting fire at the initial stages and help in saving lives and property. Commercial fire detecting systems usually have an alarm signaling with a buzzer, which has been found inadequate. To overcome these deficiencies, this project aims to design a fire detection system. In the proposed model YOLO v3 is used to detect the fire in images. Initially, it will train the dataset and after the training, the trained model is used for further process. The user has an option to upload the image and click submit, after this request hitting on the backend endpoint, identification of features will be done and soon the output will be shown saying that the image has fired or not. Also, if the fire is detected, an email and the location will also be sent to the given email address. There are quite a few such existing systems, however, they are not as effective as the situation demands to need. So a new system to detect fire from infrared images have proposed. By making use of computer vision and machine learning techniques to make it efficient and reliable. Also, the system uses brightness classification along with image processing and histogram-based segmentation. All these are made to increase the accuracy and make the system more suitable for real-time implementation. Thus, the proposed system not only solves the existing problems but also provides a new and efficient approach. Here YOLO V3 is used as a classifier to classify based on the parameters learned during the training process and the binary cross-entropy loss function is used to figure out how good our model is. Accuracy is used as an evaluation metric. Hence the main goal of the project is to accurately detect the presence of the fire as early as possible.

Optimisation of Deep Learning Small-Object Detectors with Novel Explainable Verification.

In this paper, we present a novel methodology based on machine learning for identifying the most appropriate from a set of available state-of-the-art object detectors for a given application. Our particular interest is to develop a road map for identifying verifiably optimal selections, especially for challenging applications such as detecting small objects in a mixed-size object dataset. State-of-the-art object detection systems often find the localisation of small-size objects challenging since most are usually trained on large-size objects. These contain abundant information as they occupy a large number of pixels relative to the total image size. This fact is normally exploited by the model during training and inference processes. To dissect and understand this process, our approach systematically examines detectors’ performances using two very distinct deep convolutional networks. The first is the single-stage YOLO V3 and the second is the double-stage Faster R-CNN. Specifically, our proposed method explores and visually illustrates the impact of feature extraction layers, number of anchor boxes, data augmentation, etc., utilising ideas from the field of explainable Artificial Intelligence (XAI). Our results, for example, show that multi-head YOLO V3 detectors trained using augmented data produce better performance even with a fewer number of anchor boxes. Moreover, robustness regarding the detector’s ability to explain how a specific decision was reached is investigated using different explanation techniques. Finally, two new visualisation techniques are proposed, WS-Grad and Concat-Grad, for identifying explanation cues of different detectors. These are applied to specific object detection tasks to illustrate their reliability and transparency with respect to the decision process. It is shown that the proposed techniques can result in high resolution and comprehensive heatmaps of the image areas, significantly affecting detector decisions as compared to the state-of-the-art techniques tested.

Efficient Detection Method of Pig-Posture Behavior Based on Multiple Attention Mechanism.

Due to the low detection precision and poor robustness, the traditional pig-posture and behavior detection method is difficult to apply in the complex pig captivity environment. In this regard, we designed the HE-Yolo (High-effect Yolo) model, which improves the Darknet-53 feature extraction network and integrates DAM (Dual attention mechanism) of channel attention mechanism and space attention mechanism, to recognize the posture behaviors of the enclosure pigs in real-time. First, the pig data set is clustered and optimized by the K-means algorithm to obtain a new anchor frame size. Second, the DSC (Depthwise separable convolution) and h-switch activation function are innovatively introduced into the Darknet-53 feature extraction network, and the C-Res (Contrary residual structure) unit is designed to build Darknet-A feature extraction network, so as to avoid network gradient explosion and ensure the integrity of feature information. Subsequently, DAM integrating the spatial attention mechanism and the channel attention mechanism is established, and it is further combined with the Incep-abate module to form DAB (Dual attention block), and HE-Yolo is finally built by Darknet-A and DAB. A total of 2912 images of 46 enclosure pigs are divided into the training set, the verification set, and the test set according to the ratio of 14 : 3:3, and the recognition performance of HE-Yolo is verified according to the parameters of the precision P, the recall R, the AP (i.e., the area of P-R curve) and the MAP (i.e., the average value of AP). The experiment results show that the AP values of HE-Yolo reach 99.25%, 98.41%, 94.43%, and 97.63%, respectively, in the recognition of four pig-posture behaviors of standing, sitting, prone and sidling of the test set. Compared with other models such as Yolo v3, SSD, and faster R–CNN, the mAP value of HE-Yolo is increased by 5.61%, 4.65%, and 0.57%, respectively, and the single-frame recognition time of HE-Yolo is only 0.045 s. In the recognition of images with foreign body occlusion and pig adhesion, the mAP values of HE-Yolo are increased by 4.04%, 4.94%, and 1.76%, respectively, while compared with other models. Under different lighting conditions, the mAP value of HE-Yolo is also higher than that of other models. The experimental results show that HE-Yolo can recognize the pig-posture behaviors with high precision, and it shows good generalization ability and luminance robustness, which provides technical support for the recognition of pig-posture behaviors and real-time monitoring of physiological health of the enclosure pigs.

A Combined Approach for Accurate and Accelerated Teeth Detection on Cone Beam CT Images.

Teeth detection and tooth segmentation are essential for processing Cone Beam Computed Tomography (CBCT) images. The accuracy decides the credibility of the subsequent applications, such as diagnosis, treatment plans in clinical practice or other research that is dependent on automatic dental identification. The main problems are complex noises and metal artefacts which would affect the accuracy of teeth detection and segmentation with traditional algorithms. In this study, we proposed a teeth-detection method to avoid the problems above and to accelerate the operation speed. In our method, (1) a Convolutional Neural Network (CNN) was employed to classify layer classes; (2) images were chosen to perform Region of Interest (ROI) cropping; (3) in ROI regions, we used a YOLO v3 and multi-level combined teeth detection method to locate each tooth bounding box; (4) we obtained tooth bounding boxes on all layers. We compared our method with a Faster R-CNN method which was commonly used in previous studies. The training and prediction time were shortened by 80% and 62% in our method, respectively. The Object Inclusion Ratio (OIR) metric of our method was 96.27%, while for the Faster R-CNN method, it was 91.40%. When testing images with severe noise or with different missing teeth, our method promises a stable result. In conclusion, our method of teeth detection on dental CBCT is practical and reliable for its high prediction speed and robust detection.

OBJECT DETECTION USING SEMI SUPERVISED LEARNING METHODS

Object detection is used to identify objects in real time using some deep learning algorithms. In this work, wheat plant data set around the world is collected to study the wheat heads. Using global data, a common solution for measuring the amount and size of wheat heads is formulated. YOLO V3 (You Look Only Once Version 3) and Faster RCNN is a real time object detection algorithm which is used to identify objects in videos and images. The global wheat detection dataset is used for the prediction which contains 3000+ training images and few test images with csv files which have information about the ground box labels of the images. To build a data pipeline for the model Tensorflow data API or Keras Data Generators is used.

Towards Human Action Recognition during Surgeries using De-identified Video Data

Abstract With the progress of technology in modern hospitals, an intelligent perioperative situation recognition will gain more relevance due to its potential to substantially improve surgical workflows by providing situation knowledge in real-time. Such knowledge can be extracted from image data by machine learning techniques but poses a privacy threat to the staff’s and patients’ personal data. De-identification is a possible solution for removing visual sensitive information. In this work, we developed a YOLO v3 based prototype to detect sensitive areas in the image in real-time. These are then deidentified using common image obfuscation techniques. Our approach shows that it is principle suitable for de-identifying sensitive data in OR images and contributes to a privacyrespectful way of processing in the context of situation recognition in the OR.

Graphite Classification of Gray Cast Iron in Metallographic via a Deep Learning Approach

<p>In addition to measurements of physical and mechanical properties, quality inspections also include metallographic analyses. When gray casting iron material, different manufacturing processes cause different microstructures in the material, whose metallographic images also perform large differences. The metallographic properties of gray iron can be divided into six types (from Type A to Type F). The proportion of types will influence the strength, wear resistance, and lifetime of specimens. The determination of type is usually dependent on manual judgments. In this study, two approaches were developed to analyze six metallographic types of gray casting iron. The first approach was to determine the type according to features of the detected particles in the metallographic materials by morphology algorithm. Types A, C, and F could be identified with the shape factor (SF) of gray casting iron. Then, the remained part could be identified using average grayscale values of the part-region of the metallographic material. Second approach was to identify Types A, C, and F with SF method and then identify the remaining part through the classification of the YOLO V3 deep learning algorithm. The results showed that the second approach performed more suitably in identifying the types of metallographic of gray casting iron.</p> <p> </p>

Comparative of Viola-Jones and YOLO v3 for Face Detection in Real time

This Face detection is considering one of the important topics for recognizing human, it is the first step before the face recognition process, it is considered one of the biggest challenges in the field of vision computer. In recent years Many algorithms for detection have appeared, which depend on extracting the features of the human face, and works continue to develop them to this day. This paper aims to make a comparison between two of the most commonly face detection methods, Viola Jones (V_J) and YOLO v3. This comparison is made to determine which of the two algorithms is being most useful when used to detect faces in digital video. These algorithms are used in many applications, including image classification, medical analysis of image, and objects detection in real time (especially in surveillance cameras). Both algorithms are applied to detect faces in the real time video. The experimental results of a sample consists of 20 video frames show that V_J algorithm consumes less time in comparison with YOLO v3 algorithm, but its results are less accurate, unlike the YOLO v3 algorithm, which is slower in detect face with high accurate rate. Index Terms— Viola jones, Yolo v3, V_J algorithm, Accuracy rate.

Pigeon cleaning behavior detection algorithm based on light-weight network

The behavior of pigeons in the dovecote reflects their environmental comfort and health indicators. In order to solve the problems of time-consuming, labor-consuming, and subjectivity of traditional manual experience, an improved YOLO V4 light-weight target detection algorithm was proposed for row detection of breeding pigeons. Employ SPP, FPN, and PANet networks to strengthen the features retrieved from GhostNet as the backbone. To ensure accuracy, Ghostnet-yolo V4 reduced the model's number of parameters and raised its size to 43 MB. The light-weight feature extraction network GhostNet outperformed MobileNet V1~V3 under the modified model. Faster RCNN, SSD, YOLO V4 and YOLO V3 compression rates were increased by 43.4 percent, 35.8 percent, 70.1 percent, and 69.1 percent, respectively. The improved algorithm has an accuracy of 97.06 percent and a recognition speed of 0.028 s per frame. The improved model can provide a theoretical foundation and technological reference for detecting breeding pigeon behavior in real-time in a dovecote.

AN APPROACH FOR VEHICLE TARGET DETECTION USING YOLO V3

In recent years, autonomous driving has become the key development and transformation direction of the traditional automobile industry. In the complex driving environment, automatic driving requires extremely high accuracy and high speed of detection, otherwise, it is easy to lead to traffic accident tragedy. Given the low accuracy of traditional car target detection methods in complex scenes, combined with the current hot development of deep learning, this project applies the YOLO V3 algorithm framework to achieve car target detection. In this project, we will detect cars moving on the road using deep learning techniques. Both detection and tracking vehicles such as cars, trucks, motorcycles, etc are challenging problems, especially in complex real-world scenes that commonly involve multiple vehicles, complicated occlusions, and cluttered or even moving backgrounds. This project mainly focuses on car detection can locate cars even in complex street scenes and it can apply to other vehicles by including other datasets and their class names, but false positives have remained frequent. The approximate articulation of each car is detected using YOLO V3 pretrained detector network and untrained network and their average precision and confidence scores are compared among them. The vehicle dataset having cars is used here. Software Tools: MATLAB 2021a or Above Keywords — Object Detection, Convolutional Neural Network, Deep Learning, YOLO V3