YOLO Object Detector Research Articles

Implementation of A.I. Vehicle Detection for Traffic Analysis Using In-situ Surveillance Infrastructure

Traffic flow parameters are required for optimizing traffic operations, design of pavements, and future planning of traffic networks. Unfortunately, due to the unique characteristics and variety of vehicles in the sub-continent i.e., size and design, the accuracy of results for a vision-based system is challenged, since most thorough datasets are based on European and American traffic. This paper proposes a solution by developing a detection model ground-up using a dataset created from the local traffic surveillance footage, and creating a python pipeline for vehicle speed detection and classification. The vehicle classification model is developed using the state-of-the-art YOLO object detector which significantly reduces the computation time required to maintain the efficiency of the proposed solution. Furthermore, a computer-vision script is developed to track the movement of vehicles in the footage and record the speeds in a spreadsheet. The technique used eliminates the video calibration, including distance and angle, required for detecting accurate speeds. Finally, the realtime traffic data is analyzed to derive the fundamental traffic flow parameters and discuss the relation between flow and density. To ascertain the validity of this survey technique, the results are compared to the following renowned traffic flow models: The Modified Greenberg model, Eddie’s model, and The Two-regime model. The results are found to closely follow the models in all three cases.

Utilizing High Resolution Satellite Imagery for Automated Road Infrastructure Safety Assessments.

The European Commission (EC) has published a European Union (EU) Road Safety Framework for the period 2021 to 2030 to reduce road fatalities. In addition, the EC with the EU Directive 2019/1936 requires a much more detailed recording of road attributes. Therefore, automatic detection of school routes, four classes of crosswalks, and divided carriageways were performed in this paper. The study integrated satellite imagery as a data source and the Yolo object detector. The satellite Pleiades Neo 3 with a spatial resolution of 0.3 m was used as the source for the satellite images. In addition, the study was divided into three phases: vector processing, satellite imagery processing, and training and evaluation of the You Only Look Once (Yolo) object detector. The training process was performed on 1951 images with 2515 samples, while the evaluation was performed on 651 images with 862 samples. For school zones and divided carriageways, this study achieved accuracies of 0.988 and 0.950, respectively. For crosswalks, this study also achieved similar or better results than similar work, with accuracies ranging from 0.957 to 0.988. The study also provided the standard performance measure for object recognition, mean average precision (mAP), as well as the values for the confusion matrix, precision, recall, and f1 score for each class as benchmark values for future studies.

Deep Transfer Learning-Based Animal Face Identification Model Empowered with Vision-Based Hybrid Approach

The importance of accurate livestock identification for the success of modern livestock industries cannot be overstated as it is essential for a variety of purposes, including the traceability of animals for food safety, disease control, the prevention of false livestock insurance claims, and breeding programs. Biometric identification technologies, such as thumbprint recognition, facial feature recognition, and retina pattern recognition, have been traditionally used for human identification but are now being explored for animal identification as well. Muzzle patterns, which are unique to each animal, have shown promising results as a primary biometric feature for identification in recent studies. Muzzle pattern image scanning is a widely used method in biometric identification, but there is a need to improve the efficiency of real-time image capture and identification. This study presents a novel identification approach using a state-of-the-art object detector, Yolo (v7), to automate the identification process. The proposed system consists of three stages: detection of the animal’s face and muzzle, extraction of muzzle pattern features using the SIFT algorithm and identification of the animal using the FLANN algorithm if the extracted features match those previously registered in the system. The Yolo (v7) object detector has mean average precision of 99.5% and 99.7% for face and muzzle point detection, respectively. The proposed system demonstrates the capability to accurately recognize animals using the FLANN algorithm and has the potential to be used for a range of applications, including animal security and health concerns, as well as livestock insurance. In conclusion, this study presents a promising approach for the real-time identification of livestock animals using muzzle patterns via a combination of automated detection and feature extraction algorithms.

Enhancing Information Fusion from UAV-Captured High-Altitude Infrared Imagery through Machine Learning

Unmanned aerial vehicles (UAVs) equipped with high-altitude infrared imaging have revolutionized data collection, providing better spatial and temperature resolutions. However, an effective way to fuse and interpret this multidimensional data remains a challenge. Therefore, this research tackles this issue by incorporating machine learning specifically the YOLO object detector to fuse and analyze information from UAV-captured high-altitude infrared images. The process entails a careful fusion of data, feature extraction, and model configuration that is tailored to the unique qualities of infrared imagery. Furthermore, the confabulated YOLO model performs exceptionally well in detecting and localizing objects within the thermal spectrum. Results showed precise identification of objects as well as their localization thus indicating potential for advanced aerial surveillance and monitoring. This research represents a significant advancement in situation awareness across environmental monitoring, infrastructure inspection, and disaster response among other areas hence demonstrating the transformative ability of machine learning in aerial imaging analysis.

In-Orchard Sizing of Mango Fruit: 1. Comparison of Machine Vision Based Methods for On-The-Go Estimation

Estimation of fruit size on-tree is useful for yield estimation, harvest timing and market planning. Automation of measurement of fruit size on-tree is possible using RGB-depth (RGB-D) cameras, if partly occluded fruit can be removed from consideration. An RGB-D Time of Flight camera was used in an imaging system that can be driven through an orchard. Three approaches were compared, being: (i) refined bounding box dimensions of a YOLO object detector; (ii) bounding box dimensions of an instance segmentation model (Mask R-CNN) applied to canopy images, and (iii) instance segmentation applied to extracted bounding boxes from a YOLO detection model. YOLO versions 3, 4 and 7 and their tiny variants were compared to an in-house variant, MangoYOLO, for this application, with YOLO v4-tiny adopted. Criteria developed to exclude occluded fruit by filtering based on depth, mask size, ellipse to mask area ratio and difference between refined bounding box height and ellipse major axis. The lowest root mean square error (RMSE) of 4.7 mm and 5.1 mm on the lineal length dimensions of a population (n = 104) of Honey Gold and Keitt varieties of mango fruit, respectively, and the lowest fruit exclusion rate was achieved using method (ii), while the RMSE on estimated fruit weight was 113 g on a population weight range between 180 and 1130 g. An example use is provided, with the method applied to video of an orchard row to produce a weight frequency distribution related to packing tray size.

Automated vehicle data pipeline for accident reconstruction: New insights from LiDAR, camera, and radar data

As automated vehicles are deployed across the world, it has become critically important to understand how these vehicles interact with each other, as well as with other conventional vehicles on the road. One such method to achieve a deeper understanding of the safety implications for Automated Vehicles (AVs) is to analyze instances where AVs were involved in crashes. Unfortunately, this poses a steep challenge to crash-scene investigators. It is virtually impossible to fully understand the factors that contributed to an AV involved crash without taking into account the vehicle’s perception and decision making. Furthermore, there is a tremendous amount of data that could provide insight into these crashes that is currently unused, as it also requires a deep understanding of the sensors and data management of the vehicle. To alleviate these problems, we propose a data pipeline that takes raw data from all on-board AV sensors such as LiDAR, radar, cameras, IMU’s, and GPS’s. We process this data into visual results that can be analyzed by crash scene investigators with no underlying knowledge of the vehicle’s perception system. To demonstrate the utility of this pipeline, we first analyze the latest information on AV crashes that have occurred in California and then select two crash scenarios that are analyzed in-depth using high-fidelity synthetic data generated from the automated vehicle simulator CARLA. The data visualization procedure is demonstrated on the real-world Kitti dataset by using the YOLO object detector and a monocular depth estimator called AdaBins. Depth from LIDAR is used as ground truth to calibrate and assess the effect of noise and errors in depth estimation. The visualization and data analysis from these scenarios clearly demonstrate the vast improvement in crash investigations that can be obtained from utilizing state-of-the-art sensing and perception systems used on AVs.

Implementation Counting and Yolo Object Detection Methods for Identification Degree of Road Saturation

Identification of road conditions such as congestion at this time still requires manual efforts in getting results. The congestion parameters certainly need to be monitored, especially in crowded areas and big city business centers, such as on Jalan Jendral Sudirman, Central Jakarta. Many parameters to identify the level of road congestion, one of them is by observing the value of the degree of road saturation. The purpose of this study is to propose a system that can calculate the value of the degree of saturation quickly and accurately using a camera. The method that is being proposed is to combine the Computer Vision with YOLO Object Detector techniques based on Deep Learning and the Object Counting method to get the value of traffic flow in the observed area. The results obtained by this system are quite good, this is supported by the error value obtained by the system around 3-4%.

YOLO Object Detector and Inception-V3 Convolutional Neural Network for Improved Brain Tumor Segmentation

For a range of medical analysis applications, the localization of brain tumors and brain tumor segmentation from magnetic resonance imaging (MRI) are challenging yet critical jobs. Many recent studies have included four modalities: i.e., T1, T1c, T2 & FLAIR, it is because every tumor causing area can be detailed examined by each of these brain imaging modalities. Although the BRATS 2018 datasets give impressive segmentation results, the results are still more complex and need more testing and more training. That’s why this paper recommends operated pre-processing strategies on a small part of an image except for a full image because that’s how an effective and flexible segmented system of brain tumor can be created. In the first phase, an ensemble classification model is developed using different classifiers such as decision tree, SVM, KNN etc. to classify an image into the tumor and non-tumor class by using the strategy of using a small section can completely solve the over-fitting problems and reduces the processing time in a model of YOLO object detector using inceptionv3 CNN features. The second stage is to recommend an efficient and basic Cascade CNN (C-ConvNet/C-CNN), as we deal with a tiny segment of the brain image in each and every slice. In two independent ways, the Cascade-Convolutional Neutral Network model extracts learnable features. On the dataset of BRATS 2018, BRATS 2019 and BRATS 2020, the extensive experimental task has been carried out on the proposed tumor localization framework.: the IoU score achieved of three datasets are 97%, 98% and 100%. Other qualitative evaluations & quantitative evaluations are discussed and presented in the manuscript in detail.

Cross‐scale feature fusion connection for a YOLO detector

Multi-scale feature fusion is often used to address the issue of scale variations in object detection. However, most of the proposed network architectures only combine the features of two adjacent levels sequentially, so the first fusion nodes in both top-down and bottom-up pathways must be blank nodes that only have one input with no feature fusion. In this work, cross-scale feature fusion connection (CFFC) is proposed which aims to enhance the entire feature hierarchy by propagating the features of each level more efficiently. The proposed method reuses and aggregates all the features of other scales to the blank nodes in both top-down and bottom-up pathways. Furthermore, the authors remove the 1 × 1 convolutional layer and replace the shortcut with concatenation before fusing multiple features. These concatenated feature maps are then supervised by the channel attention block at the fusion nodes. This modification allows the network to learn the important degree of each level in concatenated feature maps along the channel dimension. It is also observed that the proposed method alleviates the inconsistency in feature pyramids with fewer parameters. The performance of a YOLO object detector equipped with the proposed method on the COCO test-dev 2017 is evaluated. The results show that the proposed method outperforms other architectures presented in the literature.

An efficient and layout‐independent automatic license plate recognition system based on the YOLO detector

This paper presents an efficient and layout-independent Automatic License Plate Recognition (ALPR) system based on the state-of-the-art YOLO object detector that contains a unified approach for license plate (LP) detection and layout classification to improve the recognition results using post-processing rules. The system is conceived by evaluating and optimizing different models, aiming at achieving the best speed/accuracy trade-off at each stage. The networks are trained using images from several datasets, with the addition of various data augmentation techniques, so that they are robust under different conditions. The proposed system achieved an average end-to-end recognition rate of 96.9% across eight public datasets (from five different regions) used in the experiments, outperforming both previous works and commercial systems in the ChineseLP, OpenALPR-EU, SSIG-SegPlate and UFPR-ALPR datasets. In the other datasets, the proposed approach achieved competitive results to those attained by the baselines. Our system also achieved impressive frames per second (FPS) rates on a high-end GPU, being able to perform in real time even when there are four vehicles in the scene. An additional contribution is that we manually labeled 38,351 bounding boxes on 6,239 images from public datasets and made the annotations publicly available to the research community.

MP34-06 MACHINE LEARNING USING A MULTI-TASK CONVOLUTIONAL NEURAL NETWORKS CAN ACCURATELY ASSESS ROBOTIC SKILLS

You have accessJournal of UrologySurgical Technology & Simulation: Training & Skills Assessment I (MP34)1 Apr 2020MP34-06 MACHINE LEARNING USING A MULTI-TASK CONVOLUTIONAL NEURAL NETWORKS CAN ACCURATELY ASSESS ROBOTIC SKILLS Jeffrey Gahan*, Ryan Steinberg, Alaina Garbens, Xingming Qu, and Eric Larson Jeffrey Gahan*Jeffrey Gahan* More articles by this author , Ryan SteinbergRyan Steinberg More articles by this author , Alaina GarbensAlaina Garbens More articles by this author , Xingming QuXingming Qu More articles by this author , and Eric LarsonEric Larson More articles by this author View All Author Informationhttps://doi.org/10.1097/JU.0000000000000878.06AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: Surgical skill evaluation relies on either direct observation or video review by humans. Both are time consuming, costly and difficult to perform at a large scale. Machine learning could allow video reviewing to economically viable at a large scale. We sought to train and evaluate multi-task convolutional neural network to predict surgical proficiency score using a validated robotic assisted suturing model. METHODS: Twenty-three videos of surgeons with varying robotic skill levels completing a validated urethrovesical anastomosis model were utilized. Global evaluation assessment of robotic skills (GEARS) scoring was assigned to each video by expert reviewers and used as training data for the machine learning algorithm. The front wrist joint, rear wrist joint, and needle driver tip of each instrument, and needle, were manually labelled in 300 frames to train a YOLO object detector (Figure 1). Each video was randomly broken down into 6 second clips, providing multiple representative clips for each video. A heatmap representing object movement based on coordinate differences over time was generated (Figure 2). Using a multi-task convolutional neural network, each clip was assigned a GEARS score. A composite GEARS score for each video was then calculated using a novel video pooling method. Model training was performed using leave-one-out cross validation. The output was the predicted score for each GEARS domain. Success was defined when the model accurately predicted the domain score within the range assigned by human experts. RESULTS: Algorithm training for 23 videos of 1-2 minute duration took 4 hours using a supercomputer. The multi-task model successfully predicted 6 of 6 domains in 65% of videos, 5 of 6 domains in 74% of videos and 4 of 6 domains in 74% of videos. CONCLUSIONS: With limited training data, a novel multi-task convolutional neural network can accurately predict surgical proficiency as assessed by GEARS scoring in a urethrovesical model. Further testing using actual operative videos is warranted. Source of Funding: None © 2020 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 203Issue Supplement 4April 2020Page: e505-e505 Advertisement Copyright & Permissions© 2020 by American Urological Association Education and Research, Inc.MetricsAuthor Information Jeffrey Gahan* More articles by this author Ryan Steinberg More articles by this author Alaina Garbens More articles by this author Xingming Qu More articles by this author Eric Larson More articles by this author Expand All Advertisement PDF downloadLoading ...

Active Player Detection in Handball Scenes Based on Activity Measures.

In team sports training scenes, it is common to have many players on the court, each with his own ball performing different actions. Our goal is to detect all players in the handball court and determine the most active player who performs the given handball technique. This is a very challenging task, for which, apart from an accurate object detector, which is able to deal with complex cluttered scenes, additional information is needed to determine the active player. We propose an active player detection method that combines the Yolo object detector, activity measures, and tracking methods to detect and track active players in time. Different ways of computing player activity were considered and three activity measures are proposed based on optical flow, spatiotemporal interest points, and convolutional neural networks. For tracking, we consider the use of the Hungarian assignment algorithm and the more complex Deep SORT tracker that uses additional visual appearance features to assist the assignment process. We have proposed the evaluation measure to evaluate the performance of the proposed active player detection method. The method is successfully tested on a custom handball video dataset that was acquired in the wild and on basketball video sequences. The results are commented on and some of the typical cases and issues are shown.