Integrated Detection of Traffic Signs and Lane Changes Using Deep Learning for Autonomous Driving Systems

Abstract

This research presents a modular deep learning- based pipeline designed to integrate traffic sign detection, lane segmentation, and lane change prediction for autonomous driv- ing. ThesystemutilizesYOLOv8 forobjectdetection (trafficsigns andspeedbreakers),andYOLOv8-segforlanesegmentation. A custom logic module processes lane masks for accurate lane changeprediction,whileGoogleText-to-Speech(gTTS)generates audio alerts. The pipeline supports real-time performance with GPU acceleration and processes videos offline with visual and verbalfeedback.Resultssuggesthighprecisionindetection and practical application for advanced driver assistance systems (ADAS).

Introduction

The study presents an Autonomous Driving Assist System that integrates state-of-the-art deep learning models with rule-based logic and audio feedback to enhance real-time perception of road environments. Leveraging YOLOv8 for object detection (traffic signs and speed breakers) and YOLOv8-seg for pixel-level lane segmentation, the system accurately detects critical road elements. A custom lane change prediction algorithm analyzes lane mask movements across video frames to anticipate lane changes.

The system incorporates Google Text-to-Speech (gTTS) to provide synchronized verbal alerts, improving safety and accessibility. It supports offline video processing optimized for GPU acceleration, enabling real-time or near-real-time performance on embedded hardware.

Key components include:

• Object Detection: YOLOv8m detects multiple traffic signs and speed breakers with high accuracy.

• Lane Segmentation: YOLOv8-seg provides detailed lane and vehicle segmentation, supporting lane change prediction.

• Lane Change Prediction: A lightweight, rule-based temporal algorithm infers lane changes by tracking lane centroid shifts.

• Audio Feedback: Text-to-speech generates synchronized verbal alerts for detected events.

• Pipeline Architecture: Modular, sequential video processing stages generate annotated videos with audio alerts.

The system achieves high accuracy metrics (mAP > 97% for detection, IoU ~96% for segmentation) and low latency (~35–40 ms/frame), suitable for embedded automotive applications. It fills a research gap by combining unified multimodal perception with audio narration, enhancing driver assistance, autonomous vehicle safety, and accessibility.

Conclusion

Thisworkpresentsamodularandefficientautonomous driving assistant system that integrates YOLOv8-based object detectionandsegmentationmodelswithvoicealertmech- anismsusinggTTSandMoviePy.Thesystemeffectively handlesreal-timelanedetection,lanechangeprediction,traffic signrecognition,andspeedbreakeridentification,allwithin anoffline,command-lineenvironment.Themodulardesign ensures that each component functions independently yet col- laboratively, offeringflexibilityforupgradesandoptimization.Theresultsdemonstratethatdeeplearningandcomputervisiontechniquescanbeappliedinpractical,cost-effective waystoenhancedriverawarenessandroadsafety.High detection accuracy and real-time performance combined with intuitiveaudioalertsmakethesystemareliablefoundationforintelligenttransportationapplications. For future work, we aim to extend support to live video streams and deploy the system on edge devices such as NVIDIA Jetson and Raspberry Pi. Additional improvements includeadoptingdeepreinforcementlearningforadaptivelane changelogic,expandingtraffic signcategories, andenhancing detection under low-light and adverse weather conditions.

References

[1] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You only lookonce: Unified, real-time object detection,” Proceedings of the IEEEconference on computer vision and pattern recognition, pp. 779–788,2016. [2] G. e. a. Jocher, “Yolov8 by ultralytics.” https://github.com/ultralytics/ultralytics, 2023.Accessed: 2025-04-06. [3] Ultralytics,“Yolov8-seg:Real-timesegmentationwithyolov8,”GitHubRepository, 2023.Accessed: 2025-04-06. [4] X. Pan, J. Shi, P. Luo, X. Wang, and X. Tang, “Spatial as deep: Spatialcnn for traffic lane detection,” in Proceedings of the AAAI Conferenceon Artificial Intelligence, vol. 32, 2018. [5] J. Stallkamp, M. Schlipsing, J. Salmen, and C. Igel, “Man vs. computer:Benchmarkingmachinelearningalgorithmsfortrafficsignrecognition,”Neural networks, vol. 32, pp. 323–332, 2012. [6] P. Agrawal et al., “Speed breaker detection and alert system using gpsand accelerometer sensor,” International Journal of Computer Applica-tions, vol. 111, no. 14, 2015. [7] A. Gupta et al., “Real-time speed breaker detection and alert systemusing convolutional neural networks,” International Conference on Ma-chine Learning and Computing (ICMLC), 2019. [8] N. Deo and M. M. Trivedi, “Convolutional social pooling for vehicletrajectory prediction,” in Proceedings of the IEEE/CVF Conference onComputer Vision and Pattern Recognition Workshops, pp. 1468–1476,2018. [9] K. Chang, S. Li, Y. Xiang, Y. Lin, and B. Li, “Interpretable and rule-based lane change prediction for autonomous driving,” IEEE Transac-tions on Intelligent Vehicles, vol. 6, no. 3, pp. 471–482, 2021. [10] D.Neven,B.DeBrabandere,S.Georgoulis,M.Proesmans,andL.VanGool,“Towardsendtoendlanedetection:aninstancesegmentationapproach,”2018IEEEIntelligentVehiclesSymposium(IV),pp.286–291, 2018. [11] R. Bakht, S. S. Chowdhury, et al., “Voice-based navigation system forvisuallyimpaired peopleusing googlett sandgps,”in2020InternationalConference on Computing, Power and Communication Technologies(GUCON), pp. 291–296, IEEE, 2020. [12] J. Shen, R. Pang, R. J. Weiss, et al., “Natural tts synthesis by condition-ing wavenet on mel spectrogram predictions,” 2018 IEEE InternationalConferenceonAcoustics,SpeechandSignalProcessing(ICASSP),pp.4779–4783,2018. [13] M. R. Bhuiyan, M. Rahman, et al., “Ai-powered assistance for thevisuallyimpairedusingobjectdetectionandtext-to-speech,”in20212ndInternational Conference on Robotics, Electrical and Signal ProcessingTechniques (ICREST), pp. 295–300, IEEE, 2021.

Copyright

Copyright © 2025 Atharav Ghadge, Manas Manekar, Pushkar Mhatre, Palak Pandey, Prof. Kirti Randhe, Milind Ankleshwar. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.