Optimizing YOLO Performance for Traffic Light Detection and End-to-End Steering Control for Autonomous Vehicles in Gazebo-ROS2

International Journal of Advanced Computer Science and Applications, Journal Year: 2023, Volume and Issue: 14(7)

Published: Jan. 1, 2023

Autonomous driving has become a popular area of research in recent years, with accurate perception and recognition the environment being critical for successful implementation. Traditional methods recognizing controlling steering rely on color shape traffic lights road lanes, which can limit their ability to handle complex scenarios variations data. This paper presents an optimization You Only Look Once (YOLO) object detection algorithm light end-to-end control lane-keeping simulation environment. The study compares performance YOLOv5, YOLOv6, YOLOv7, YOLOv8 models signal detection, achieving best results mean Average Precision (mAP) 98.5%. Additionally, proposes convolutional neural network (CNN) based angle controller that combines data from classical proportional integral derivative (PID) human perception. predicts accurately, outperforming conventional open-source computer vision (OpenCV) methods. proposed algorithms are validated autonomous vehicle model simulated Gazebo Robot Operating System 2 (ROS2).

Language: Английский

---

AInsectID Version 1.1: An Insect Species Identification Software Based on the Transfer Learning of Deep Convolutional Neural Networks

Published: March 25, 2025

AInsectID Version 1.1 is a Graphical User Interface (GUI)‐operable open‐source insect species identification, color processing, and image analysis software. The software has current database of 150 insects integrates artificial intelligence approaches to streamline the process with focus on addressing prediction challenges posed by mimics. This paper presents methods algorithmic development, coupled rigorous machine training used enable high levels validation accuracy. Our work transfer learning prominent convolutional neural network (CNN) architectures, including VGG16, GoogLeNet, InceptionV3, MobileNetV2, ResNet50, ResNet101. Here, we employ both fine tuning hyperparameter optimization improve performance. After extensive computational experimentation, ResNet101 evidenced as being most effective CNN model, achieving accuracy 99.65%. dataset utilized for sourced from National Museum Scotland, Natural History London, open source datasets Zenodo (CERN's Data Center), ensuring diverse comprehensive collection species.

Language: Английский

---

Plant Disease Classification Using Deep Learning and the Hyperband Strategy

Lecture notes in networks and systems, Journal Year: 2025, Volume and Issue: unknown, P. 709 - 717

Published: Jan. 1, 2025

Language: Английский

---

Deep Learning-Based Lane-Keeping Assist System for Self-Driving Cars Using Transfer Learning and Fine Tuning

Journal of Advances in Information Technology, Journal Year: 2024, Volume and Issue: 15(3), P. 322 - 329

Published: Jan. 1, 2024

This paper presents an advanced lane-keeping assistance system specifically designed for self-driving cars.The proposed model combines the powerful Xception network with transfer learning and fine-tuning techniques to accurately predict steering angle.By analyzing cameracaptured images, effectively learns from human driving knowledge provides precise estimations of angle necessary safe lane-keeping.The technique allows leverage extensive acquired ImageNet dataset, while is utilized tailor pre-trained specific task prediction based on input enabling optimal performance.Fine-tuning was initiated by initially freezing training only Fully Connected (FC) layer first 10 epochs.Subsequently, entire model, encompassing both backbone FC layer, unfrozen further training.To evaluate system's effectiveness, a comprehensive comparative analysis conducted against popular existing models, including Nvidia, MobilenetV2, VGG19, InceptionV3.The evaluation includes assessment operational accuracy loss function, utilizing Mean Squared Error (MSE) equation.The achieves lowest function values validation, demonstrating its superior predictive performance.Additionally, model's performance evaluated through real-world testing pre-designed trajectories maps, resulting in minimal deviation desired trajectory over time.This practical valuable insights into mode's reliability potential assist lanekeeping tasks.

Language: Английский

---

Lane Road Segmentation Based on Improved UNet Architecture for Autonomous Driving

International Journal of Advanced Computer Science and Applications, Journal Year: 2023, Volume and Issue: 14(7)

Published: Jan. 1, 2023

This paper introduces a real-time workflow for implementing neural networks in the context of autonomous driving. The UNet architecture is specifically selected road segmentation due to its strong performance and low complexity. To further improve model's capabilities, Local Binary Convolution (LBC) incorporated into skip connections, enhancing feature extraction, elevating Intersection over Union (IoU) metric. evaluation model focuses on detection, utilizing IOU Two datasets are used training validation: widely KITTI dataset custom collected within ROS2 environment. Simulation validation performed both assess our model. demonstrates an impressive IoU score 97.90% segmentation. Moreover, when evaluated dataset, achieves 98.88%, which comparable conventional models. Our proposed method reconstruct structure provide input extraction can effectively existing lane methods.

Language: Английский

---

Investigation on Driver Drowsiness Detection using Deep Learning Approaches

Published: Aug. 10, 2023

Language: Английский

---

Steering Angle Prediction for Autonomous Vehicles Using Deep Transfer Learning

Journal of Advances in Information Technology, Journal Year: 2024, Volume and Issue: 15(1), P. 138 - 146

Published: Jan. 1, 2024

Self-driving cars are anticipated to revolutionize future transportation due their reliability, safety, and continuous learning capabilities.Researchers actively engaged in developing autonomous driving systems, employing techniques like behavioral cloning reinforcement learning.This study introduces a distinctive perspective by an end-to-end approach, using camera inputs predict steering angles based on model learned from human expertise.The demonstrates rapid training achieves over 90.1% accuracy Mean Percentage of Prediction (MPP).In this context, the aims replicate driver behavior applying transfer pre-trained VGG19 with various activation functions.The proposed is trained analyze road images as input, predicting optimal adjustments.Evaluation encompasses dataset ROS2 simulation environment, comparing results several Convolutional Neural Networks (CNNs) models including Nvidia's, MobileNet-V2, ResNet50, VGG16, VGG19.The impact functions Exponential Linear Unit (ELU), Rectified (ReLU), Leaky ReLU also explored.This research contributes advancing systems addressing real-world complexities facilitating integration into everyday transportation.The novel approach utilizing comprehensive evaluation underscores its significance optimizing self-driving technology.

Language: Английский

---

Optimizing Grape Leaf Disease Identification Through Transfer Learning and Hyperparameter Tuning

International Journal of Advanced Computer Science and Applications, Journal Year: 2024, Volume and Issue: 15(2)

Published: Jan. 1, 2024

Grapes are a globally cultivated fruit with significant economic and nutritional value, but they susceptible to diseases that can harm crop quality yield. Identifying grape leaf accurately promptly is vital for effective disease management sustainable viticulture. To address this challenge, we employ transfer learning approach, utilizing well-established pre-trained models such as ResNet50V2, ResNet152V2, MobileNetV2, Xception, In-ceptionV3, renowned their exceptional performance across various tasks. Our primary objective identify the most suitable network architecture classification of diseases. This achieved through rigorous evaluation process considers key metrics accuracy, F1 score, precision, recall, loss. By systematically assessing these models, aim select one demonstrates best on our dataset. Following model selection, proceed crucial phase fine-tuning model’s hyperparameters. essential enhance predictive capabilities overall effectiveness in identification. accomplish this, conduct an extensive hyperparameter search using Hyperband strategy. Hyperparameters play pivotal role shaping behavior deep by exploring wide range combinations, goal optimal configuration maximizes given Additionally, study’s results were compared those numerous relevant studies.

Language: Английский

---

Advanced Night time Object Detection in Driver-Assistance Systems using Thermal Vision and YOLOv5

International Journal of Advanced Computer Science and Applications, Journal Year: 2023, Volume and Issue: 14(6)

Published: Jan. 1, 2023

Driver-assistance systems have become an indispensable component of modern vehicles, serving as a crucial element in enhancing safety for both drivers and passengers. Among the fundamental aspects these systems, object detection stands out, posing significant challenges low-light scenarios, particularly during nighttime. In this research paper, we propose innovative advanced approach detecting objects nighttime driver-assistance systems. Our proposed method leverages thermal vision incorporates You Only Look Once version 5 (YOLOv5), which demonstrates promising results. The primary objective study is to comprehensively evaluate performance our model, utilizes combination stochastic gradient descent (SGD) Adam optimizer. Moreover, explore impact different activation functions, including SiLU, ReLU, Tanh, LeakyReLU, Hardswish, on efficiency within driver assistance system that imaging. To assess effectiveness employ standard evaluation metrics precision, recall, mean average precision (mAP), commonly used

Language: Английский

---

An Improved Lane-Keeping Controller for Autonomous Vehicles Leveraging an Integrated CNN-LSTM Approach

International Journal of Advanced Computer Science and Applications, Journal Year: 2023, Volume and Issue: 14(7)

Published: Jan. 1, 2023

Representing the task of navigating a car through traffic using traditional algorithms is complex endeavor that presents significant challenges. To overcome this, researchers have started training artificial neural networks data from front-facing cameras, combined with corresponding steering angles. However, many current solutions focus solely on visual information camera frames, overlooking important temporal relationships between these frames. This paper introduces novel approach to end-to-end control by combining VGG16 convolutional network (CNN) architecture Long Short-Term Memory (LSTM). integrated model enables learning both dependencies within sequence images and dynamics process. Furthermore, we will present evaluate estimated accuracy proposed for angle prediction, comparing it various CNN models including Nvidia classic model, MobilenetV2 when LSTM. The method demonstrates superior compared other approaches, achieving lowest loss function. its performance, recorded video saved results based human perception robot operating system (ROS2). videos are then split into image sequences be smoothly fed processing training.

Language: Английский

---