[论文解读] Real-Time Pothole Detection Using Deep Learning

Anas Al Shaghouri, Rami Alkhatib+1 more|arXiv (Cornell University)|Jul 13, 2021

Infrastructure Maintenance and Monitoring被引用 31

一句话总结

本文评估了多种深度学习目标检测器用于实时坑洞检测，指出 YOLOv4Darknet53 是最佳模型，具备 81% 的召回率、85% 的精确率、85.39% 的 mAP、约 20 FPS，检测范围可达 100 米。

ABSTRACT

Roads are connecting line between different places, and used daily. Roads' periodic maintenance keeps them safe and functional. Detecting and reporting the existence of potholes to responsible departments can help in eliminating them. This study deployed and tested on different deep learning architecture to detect potholes. The images used for training were collected by cellphone mounted on the windshield of the car, in addition to many images downloaded from the internet to increase the size and variability of the database. Second, various object detection algorithms are employed and compared to detect potholes in real-time like SDD-TensorFlow, YOLOv3Darknet53 and YOLOv4Darknet53. YOLOv4 achieved the best performance with 81% recall, 85% precision and 85.39% mean Average Precision (mAP). The speed of processing was 20 frame per second. The system was able to detect potholes from a range on 100 meters away from the camera. The system can increase the safety of drivers and improve the performance of self-driving cars by detecting pothole time ahead.

研究动机与目标

• Motivate automatic pothole detection to improve road safety and maintenance efficiency.
• assess whether deep learning detectors can operate in real time on road imagery from a vehicle-mounted camera.
• compare multiple detection architectures to identify a best-performing model for pothole detection.

提出的方法

• Collect pothole images from cellphone-mounted cameras and internet sources to create a diverse training dataset.
• Evaluate several object detection architectures (e.g., SSD-TensorFlow, YOLOv3Darknet53, YOLOv4Darknet53) for pothole detection.
• Measure performance in terms of recall, precision, mean Average Precision (mAP), and processing speed (frames per second).
• Determine the practical detection range of the system (e.g., potholes detectable at 100 meters).

实验结果

研究问题

• RQ1 Which deep learning object detectors provide the best accuracy for pothole detection in real-time settings?
• RQ2 What is the trade-off between detection accuracy and speed for pothole detection on vehicle-mounted systems?
• RQ3 What is the maximum detection range at which potholes can be reliably detected by the system?

主要发现

• YOLOv4Darknet53 achieved the best overall performance among the tested detectors.
• Recall: 81% and Precision: 85% for the best model.
• Mean Average Precision (mAP) of 85.39% achieved.
• Processing speed around 20 frames per second.
• Potholes detectable from a range of up to 100 meters.
• The approach can enhance driver safety and support autonomous driving systems by early pothole detection.