Lane Line Detection and Object Scene Segmentation Using Otsu Thresholding and the Fast Hough Transform for Intelligent Vehicles in Complex Road Conditions

Electronics, Journal Year: 2023, Volume and Issue: 12(5), P. 1079 - 1079

Published: Feb. 21, 2023

An Otsu-threshold- and Canny-edge-detection-based fast Hough transform (FHT) approach to lane detection was proposed improve the accuracy of for autonomous vehicle driving. During last two decades, vehicles have become very popular, it is constructive avoid traffic accidents due human mistakes. The new generation needs automatic intelligence. One essential functions a cutting-edge automobile system detection. This study recommended idea through improved (extended) Canny edge using transform. Gaussian blur filter used smooth out image reduce noise, which could help accuracy. operator known as Sobel calculated gradient intensity identify edges in an convolutional kernel. These techniques were applied initial module enhance characteristics road lanes, making easier detect them image. then routes based on mathematical relationship between lanes vehicle. It did this by converting into polar coordinate looking lines within specific range contrasting points. allowed algorithm distinguish other features After this, detection, possible left right marking extraction; region interest (ROI) must be extracted traditional approaches work effectively easily. methodology tested several sequences. least-squares fitting track lane. demonstrated high experiments, demonstrating that identification method performed well regarding reasoning speed accuracy, considered both real-time processing satisfy requirements recognition lightweight driving systems.

Language: Английский

---

Object Detection, Recognition, and Tracking Algorithms for ADASs—A Study on Recent Trends

Sensors, Journal Year: 2023, Volume and Issue: 24(1), P. 249 - 249

Published: Dec. 31, 2023

Advanced driver assistance systems (ADASs) are becoming increasingly common in modern-day vehicles, as they not only improve safety and reduce accidents but also aid smoother easier driving. ADASs rely on a variety of sensors such cameras, radars, lidars, combination sensors, to perceive their surroundings identify track objects the road. The key components object detection, recognition, tracking algorithms that allow vehicles other road, pedestrians, cyclists, obstacles, traffic signs, lights, etc. This information is then used warn potential hazards or by ADAS itself take corrective actions avoid an accident. paper provides review prominent state-of-the-art different functionalities ADASs. begins introducing history fundamentals followed reviewing recent trends various functionalities, along with datasets employed. concludes discussing future for discusses need more research challenging environments, those low visibility high density.

Language: Английский

---

Advances in Instance Segmentation: Technologies, Metrics and Applications in Computer Vision

Neurocomputing, Journal Year: 2025, Volume and Issue: unknown, P. 129584 - 129584

Published: Jan. 1, 2025

Language: Английский

---

A Novel Solution to the Real-Time Lane Detection and Tracking Problem for Autonomous Vehicles by Using Faster R-CNN and Mask R-CNN

International Journal of Automotive Science And Technology, Journal Year: 2025, Volume and Issue: 9(1), P. 71 - 80

Published: Jan. 17, 2025

Autonomous vehicle technology has advanced in the automobile sector. aims to make driving safer and reduce driver-caused traffic accidents. work toward this. Lane detection tracking are crucial autonomous systems. Mostly image processing techniques mainly utilized for lane literature. But, while performing tracking, two basic problems encountered. First one is also needs with a specific area on load correct area. The region of interest (ROI) process often used filter be worked from image. However, since fixed coordinates provided this operation, restricts oper-ation areas where it must rotated. Second weather conditions very effective lanes by utilizing techniques. There serious cloudy, sunny or momentary changes air. This study uses deep learning methods against these problems. Using Mask R-CNN faster algorithms together, eliminated successfully implemented. problem solved been tested experimentally developed tool. Both originally da-taset KITTI dataset were separately model training carried out experimental tests. systems well according

Language: Английский

---

Enhanced SCNN-Based Hybrid Spatial-Temporal Lane Detection Model for Intelligent Transportation Systems

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 40075 - 40091

Published: Jan. 1, 2024

Accurate and timely lane detection is imperative for the seamless operation of autonomous driving systems. In this study, leveraging gradual variation features within a defined range width length, we introduce an enhanced Spatial-Temporal Recurrent Neural Network (SCNN) framework. This framework serves as cornerstone innovative hybrid spatial-temporal model detection, which tailored to address prevalent issues substandard performance insufficient real-time processing in intricate scenarios, such those involving erosion inconsistent lighting conditions, often challenge conventional models. With foundational understanding that lanes manifest continuous lines, employ temporal sequence imagery input our model, thereby ensuring rich provision feature information. The adopts encoder-decoder structure integrates module extraction interrelated information from image sequence. culminates output results terminal frame. proposed exhibits commendable synthesis accuracy efficiency, attaining Accuracy 97.87%, xmlns:xlink="http://www.w3.org/1999/xlink">F ₁ -score 0.943, xmlns:xlink="http://www.w3.org/1999/xlink">FPS 19.342 on tvtLANE dataset 98.21%, 0.957 Tusimple dataset. These metrics signify superior over majority current methods.

Language: Английский

---

Performance and Robustness of Regional Image Segmentation Driven by Selected Evolutionary and Genetic Algorithms: Study on MR Articular Cartilage Images

Sensors, Journal Year: 2022, Volume and Issue: 22(17), P. 6335 - 6335

Published: Aug. 23, 2022

The analysis and segmentation of articular cartilage magnetic resonance (MR) images belongs to one the most commonly routine tasks in diagnostics musculoskeletal system knee area. Conventional regional methods, which are based either on histogram partitioning (e.g., Otsu method) or clustering methods K-means), have been frequently used for task segmentation. Such well known as fast working environment, where image features reliably recognizable. well-known fact is that performance these prone noise artefacts. In this context, strategies, driven by genetic algorithms selected evolutionary computing potential overcome traditional such thresholding K-means context their performance. These optimization strategies consecutively generate a pyramid possible set thresholds, quality evaluated using fitness function Kapur's entropy maximization find optimal combination thresholds On other hand, often computationally demanding, limitation stack MR images. study, we publish comprehensive fuzzy soft segmentation, artificial bee colony (ABC), particle swarm (PSO), Darwinian (DPSO), algorithm an selection against segmentations extraction from This study objectively analyzes upon variable with dynamic intensities report segmentation's robustness various conditions number classes (4, 7, 10), (area, perimeter, skeleton) preciseness lastly time, represents important factor We use same settings individual strategies: 100 iterations 50 population. suggests ABC gives best comparison view influence additive influence, also extraction. some cases does not give good cases, analyzed significantly except normally lower algorithms. statistical tests significance, showing differences method. Lastly, part software integrating all study.

Language: Английский

---

Design and Implementation of Lane Detection using Hough Transformation

Published: May 4, 2023

Typically, road lanes are solid or dashed line formations that continuous on the surface. As driving sceneries as well substantially overlap, placement of in one frame is highly correlated with their position next frame. Computer vision-related machine learning algorithms have also advanced rapidly recent years, becoming both more efficient & effective high-precision optical and electronic sensors become commonplace, real-time scene recognition feasible. Recent years seen several technical breakthroughs field safety, number accidents has risen at an alarming pace, driver inattention being primary causes. To minimize incidence maintain technological advances required. Lane Detection Systems, which operate to recognize lane boundaries alert if he changes goes incorrect markings, method achieving this goal. A detection system a crucial element many technologically transportation systems. This research uses Hough Transform technique for identification.

Language: Английский

---

Deep Learning-Based AI model for Road Detection

Published: Jan. 18, 2024

In the rapidly advancing domain of autonomous vehicles, ensuring robust and realtime lane detection is pivotal for safe navigation. This paper addresses challenges existing methods, marked by computational inefficiencies limited adaptability to changing configurations. Our innovative approach treats as an instance segmentation problem, utilizing LaneNet architecture incorporating metric learning enhance model's understanding features. A crucial contribution lies in integrating H-Net homography prediction during forward pass, dynamically adjusting changes road plane geometry, fitting. overcomes limitations fixed matrices traditional methods. DBSCAN clustering facilitate effective grouping, particularly scenarios with variable numbers complex changes. Tested on TuSimple dataset comprising 6,408 images, our results demonstrate superior performance compared conventional approaches. The approach's diverse scenarios, including curved lanes ground variations, positions it a significant advancement open literature. As automotive industry leans towards solutions, methodology stands poised contribute evolution precise efficient systems.

Language: Английский

---

QuantLaneNet: A 640-FPS and 34-GOPS/W FPGA-Based CNN Accelerator for Lane Detection

Sensors, Journal Year: 2023, Volume and Issue: 23(15), P. 6661 - 6661

Published: July 25, 2023

Lane detection is one of the most fundamental problems in rapidly developing field autonomous vehicles. With dramatic growth deep learning recent years, many models have achieved a high accuracy for this task. However, existing deep-learning methods lane face two main problems. First, early studies usually follow segmentation approach, which requires much post-processing to extract necessary geometric information about lines. Second, fail reach real-time speed due complexity model architecture. To offer solution these problems, paper proposes lightweight convolutional neural network that only small arrays minimum post-processing, instead maps task detection. This proposed utilizes simple representation format its output. The can achieve 93.53% on TuSimple dataset. A hardware accelerator and implemented Virtex-7 VC707 FPGA platform optimize processing time power consumption. Several techniques, including data quantization reduce width down 8-bit, exploring various loop-unrolling strategies different convolution layers, pipelined computation across are optimized implementation process at 640 FPS while consuming 10.309 W, equating throughput 345.6 GOPS energy efficiency 33.52 GOPS/W.

Language: Английский

---

Method for Autonomous Lane Detection in Assisted Driving

Published: May 23, 2023

This paper presents a machine learning-based method for detecting lanes on roads. The proposed approach includes several processing steps such as preprocessing of the original image frames, application Hough Line Transform an initial detection lanes, computation vanishing point to determine horizon line, and region interest (ROI) determination. Additionally, compensates unknown position camera facing road by cropping triangle-shaped perspective area. To correct errors caused discoloration cracks, color mask white yellow pixels is used. orientation determined analyzing slope lines, lane coordinates are linked center. uses U-Net neural network implementation based Python programming language OpenCV library. In final section we also present comparison with convolutional networks discuss results.

Language: Английский

---