Vehicle Flow Detection and Tracking Based on an Improved YOLOv8n and ByteTrack Framework DOI Creative Commons
Jinjiang Liu, Yonghua Xie, Yanwen Zhang

et al.

World Electric Vehicle Journal, Journal Year: 2024, Volume and Issue: 16(1), P. 13 - 13

Published: Dec. 28, 2024

Vehicle flow detection and tracking are crucial components of intelligent transportation systems. However, traditional methods often struggle with challenges such as the poor small objects low efficiency when processing large-scale data. To address these issues, this paper proposes a vehicle method that integrates an improved YOLOv8n model ByteTrack algorithm. In module, we introduce innovative MSN-YOLO model, which combines C2f_MLCA Detect_SEAM NWD loss function to enhance feature fusion improve cross-scale information processing. These enhancements significantly boost model’s ability detect handle complex backgrounds. incorporate algorithm train unique re-identification (Re-ID) features, ensuring robust multi-object in environments improving stability accuracy tracking. The experimental results demonstrate proposed achieves mean Average Precision (mAP) 62.8% at IoU = 0.50 Multiple Object Tracking Accuracy (MOTA) 72.16% real-time improvements represent increases 2.7% 3.16%, respectively, compared baseline algorithms. This provides effective technical support for traffic management, monitoring, congestion prediction.

Language: Английский

Multi-Agent Reinforcement Learning for task allocation in the Internet of Vehicles: Exploring benefits and paving the future DOI
Inam Ullah, Sushil Kumar Singh, Deepak Adhikari

et al.

Swarm and Evolutionary Computation, Journal Year: 2025, Volume and Issue: 94, P. 101878 - 101878

Published: Feb. 20, 2025

Language: Английский

Citations

0
Vehicle Recognition and Driving Information Detection with UAV Video Based on Improved YOLOv5-DeepSORT Algorithm DOI Creative Commons
Binshuang Zheng, Jing Zhou, Zhengqiang Hong

et al.

Sensors, Journal Year: 2025, Volume and Issue: 25(9), P. 2788 - 2788

Published: April 28, 2025

To investigate whether the skid resistance of ramp meets requirements vehicle driving safety and stability, simulation using ideal driver model is inaccurate. Therefore, considering driver’s habits, this paper proposes use Unmanned aerial vehicles (UAVs) for collection extraction information. process collected UAV video, Google Collaboration platform used to modify compile “You Only Look Once” version 5 (YOLOv5) algorithm with Python 3.7.12, YOLOv5 retrained captured video. The results show that precision rate P recall R have satisfactory an F1 value 0.86, reflecting a good P-R relationship. loss function also stabilized at very low level after 70 training epochs. Then, trained replace Faster R-CNN detector in DeepSORT improve detection accuracy speed extract information from perspective UAV. By coding, coordinate trajectory extracted, smoothed, frame difference method calculate real-time information, which convenient establishment real model.

Language: Английский

Citations

0
Drone Swarm for Distributed Video Surveillance of Roads and Car Tracking DOI Creative Commons
David Sánchez Pedroche, Daniel Amigo, Jesús Garcı́a

et al.

Drones, Journal Year: 2024, Volume and Issue: 8(11), P. 695 - 695

Published: Nov. 20, 2024

This study proposes a swarm-based Unmanned Aerial Vehicle (UAV) system designed for surveillance tasks, specifically detecting and tracking ground vehicles. The proposal is to assess how consisting of multiple cooperating UAVs can enhance performance by utilizing fast detection algorithms. Within the study, differences in one-stage two-stage models have been considered, revealing that while offer improved accuracy, their increased computation time renders them impractical real-time applications. Consequently, faster models, such as tested YOLOv8 architectures, appear be more viable option operations. Notably, approach enables these algorithms achieve an accuracy level comparable slower models. Overall, experimentation analysis demonstrates larger YOLO architectures exhibit longer processing times exchange superior success rates. However, inclusion additional introduced outweighed choice algorithm if mission correctly configured, thus demonstrating facilitates use maintaining levels alternatives. perspectives provided included hold significance, they are essential achieving enhanced results.

Language: Английский

Citations

3
Vehicle recognition pipeline via DeepSort on aerial image datasets DOI Creative Commons
Muhammad Hanzla, Muhammad Ovais Yusuf,

Naif Al Mudawi

et al.

Frontiers in Neurorobotics, Journal Year: 2024, Volume and Issue: 18

Published: Aug. 16, 2024

Introduction Unmanned aerial vehicles (UAVs) are widely used in various computer vision applications, especially intelligent traffic monitoring, as they agile and simplify operations while boosting efficiency. However, automating these procedures is still a significant challenge due to the difficulty of extracting foreground (vehicle) information from complex scenes. Methods This paper presents unique method for autonomous vehicle surveillance that uses FCM segment images. YOLOv8, which known its ability detect tiny objects, then vehicles. Additionally, system utilizes ORB features employed support recognition, assignment, recovery across picture frames. Vehicle tracking accomplished using DeepSORT, elegantly combines Kalman filtering with deep learning achieve precise results. Results Our proposed model demonstrates remarkable performance identification precision 0.86 0.84 on VEDAI SRTID datasets, respectively, detection. Discussion For tracking, achieves accuracies 0.89 0.85 respectively.

Language: Английский

Citations

2
A dataset of drilling site object detection in underground coal mines DOI
Wei Zhou, Lihong Dong, Ou Ye

et al.

China Scientific Data, Journal Year: 2024, Volume and Issue: 9(2), P. 1 - 10

Published: Jan. 1, 2024

Drilling in underground coal mine is an important measure for dealing with gas, water and hidden geological disasters, which can significantly enhance the effectiveness of disaster prevention control mining operations. In order to monitor drilling process real time improve efficiency, it necessary carry out object detection identify locate key targets at site. Compared traditional method, deep learning-based method accuracy, timeliness stability detection, but requires high-quality datasets perform well. At present, research on sites mainly relies small-scale private datasets, are insufficient providing or reliable data neural network model training. this study, we constructed a dataset site using photos taken by intrinsic safety law enforcement recorders. This developed through several steps, including cleaning, labeling, expert sampling verification. The mainstream YOLO series used quality assessment. comprises 70,948 images from under different environmental conditions, covering five categories objects: gripper, chuck, miner, helmet, drill pipe. It provides annotated files PASCAL VOC format. provide strong support sites, plays role promoting intelligent monitoring early warning.

Language: Английский

Citations

1
On-Line Detection Method of Salted Egg Yolks with Impurities Based on Improved YOLOv7 Combined with DeepSORT DOI Creative Commons

Dongjun Gong,

Shida Zhao,

Shucai Wang

et al.

Foods, Journal Year: 2024, Volume and Issue: 13(16), P. 2562 - 2562

Published: Aug. 16, 2024

Salted duck egg yolk, a key ingredient in various specialty foods China, frequently contains broken eggshell fragments embedded the yolk due to high-speed shell-breaking processes, which pose significant food safety risks. This paper presents an online detection method, YOLOv7-SEY-DeepSORT (salted SEY), designed integrate enhanced YOLOv7 with DeepSORT for real-time and accurate identification of salted yolks impurities on production lines. The proposed method utilizes as core network, incorporating multiple Coordinate Attention (CA) modules its Neck section enhance extraction subtle impurities. To address impact imbalanced sample proportions accuracy, Focal-EIoU loss function is employed, adaptively adjusting bounding box values ensure precise localization images. backbone network replaced lightweight MobileOne neural reduce model parameters improve performance. used matching tracking targets across frames, accommodating rotational variations. Experimental results demonstrate that achieves mean average precision (mAP) 0.931, reflecting 0.53% improvement over original YOLOv7. also shows performance, Multiple Object Tracking Accuracy (MOTA) Precision (MOTP) scores 87.9% 73.8%, respectively, representing increases 17.0% 9.8% SORT 2.9% 4.7% Tracktor. Overall, balances high accuracy surpassing other mainstream object methods comprehensive Thus, it provides robust solution rapid defective offers technical foundation reference future research automated safe processing products.

Language: Английский

Citations

1
Optimal Layout Method for Roadside LiDAR and Camera DOI Creative Commons
Yan Li, Han Zhang, Zhiheng Cheng

et al.

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 26905 - 26918

Published: Jan. 1, 2024

With the widespread application of LiDAR and camera, integration camera has become an urgent issue. In this study, we proposed a optimal layout method for roadside camera. Firstly, experimental design phase took into consideration various scenarios, such as curved road sections gradient sections. Secondly, video data point cloud collected from different setups were subjected to object detection recognition using YOLOv5s weights PointPillars weights, respectively. These are applied under schemes output mAP value each scheme. By comparing values schemes, scheme scene is determined. Thirdly, four parameters six installation all scenarios database. Furthermore, five machine learning algorithms employed selection. Finally, three regression with highest accuracy selected final prediction model based on control groups. Through field experiment, results show that optimized can significantly reduce blind spot vehicle occlusion problems LiDAR. The deployment increase Mean Average Precision (MAP) by over 4% through adjusting parameters. algorithm used predict cameras in unknown 95% accuracy. This improves devices vehicles changing provides guidance future

Language: Английский

Citations

0
RETRACTED: Pedestrian tracking method based on S-YOFEO framework in complex scene DOI
Wenshun Sheng, Jiahui Shen, Qiming Huang

et al.

Journal of Intelligent & Fuzzy Systems, Journal Year: 2024, Volume and Issue: unknown, P. 1 - 11

Published: March 22, 2024

This article has been retracted. A retraction notice can be found at https://doi.org/10.3233/JIFS-219433.

Language: Английский

Citations

0
Research on recognition and localization method of maize weeding robot based on improved YOLOv5 DOI Creative Commons

Lijun Zhao,

Yunfan Jia,

Wenke Yin

et al.

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: Aug. 30, 2024

Abstract In response to the challenge posed by low recognition accuracy in rugged terrains with diverse topography as well feature agricultural settings. This paper presents an optimized version of YOLOv5 algorithm alongside development a specialized laser weeding experimental platform designed for precise identification corn seedlings and weeds. The enhanced integrates effective channel attention (CBAM) mechanism while incorporating DeepSort tracking reduce parameter count seamless mobile deployment. Ablation test validate our model's achievement 96.2% along superior mAP values compared standard margins 3.1% 0.7%, respectively. Additionally, three distinct datasets capturing varied scenarios were curated; their amalgamation resulted impressive rate reaching up 96.13%. Through comparative assessments against YOLOv8, model demonstrates lightweight performance improvements including notable enhancement 2.1% coupled marginal increase 0.2% value, thus ensuring heightened precisionand robustness during dynamic object detection within intricate backgrounds.

Language: Английский

Citations

0
Study on Nighttime Pedestrian Trajectory-Tracking from the Perspective of Driving Blind Spots DOI Open Access
Wei Zhao,

Congcong Ren,

Ao Tan

et al.

Electronics, Journal Year: 2024, Volume and Issue: 13(17), P. 3460 - 3460

Published: Aug. 31, 2024

With the acceleration of urbanization and growing demand for traffic safety, developing intelligent systems capable accurately recognizing tracking pedestrian trajectories at night or under low-light conditions has become a research focus in field transportation. This study aims to improve accuracy real-time performance nighttime pedestrian-detection -tracking. A method that integrates multi-object detection algorithm YOLOP with DeepSORT is proposed. The improved incorporates C2f-faster structure Backbone Neck sections, enhancing feature extraction capabilities. Additionally, BiFormer attention mechanism introduced on recognition small-area features, CARAFE module added shallow fusion, DyHead dynamic target-detection head employed comprehensive fusion. In terms tracking, ShuffleNetV2 lightweight integrated reduce model parameters network complexity. Experimental results demonstrate proposed FBCD-YOLOP improves lane by 5.1%, increases IoU metric 0.8%, enhances speed 25 FPS compared baseline model. reached 89.6%, representing improvements 1.3%, 0.9%, 3.8% over single-task YOLO v5, multi-task TDL-YOLO, original models, respectively. These enhancements significantly model’s complex environments. enhanced achieved an MOTA 86.3% MOTP 84.9%, ID switch occurrences reduced 5. Compared ByteTrack StrongSORT algorithms, 2.9% 0.4%, were 63.6%, -tracking, making it highly suitable deployment edge computing surveillance platforms.

Language: Английский

Citations

0