VelociApp — Desktop Real-Time Road Damage Detection (YOLOv11L)

VelociApp is a desktop application for real-time road damage detection using YOLOv11L object detection. It processes live camera/video input and displays detections with bounding boxes and confidence scores, enabling faster and more consistent visual inspection of road conditions.

Key Features

• Desktop-based interface for real-time detection
• Supports video input / live feed with on-screen results
• Displays bounding boxes + confidence score for each detection
• Detects 7 road damage categories (see below)

Road Damage Categories (7)

7-Class Overview

UI Preview

Below are snapshots of the desktop interface to provide a quick overview of the application flow.

Home Screen

Mode Selection

Model Selection

Multiple YOLO model sizes were evaluated (YOLOv11m and YOLOv11L). While smaller variants can be faster, they were less consistent for real road scenes in this project. YOLOv11L was selected for better overall detection performance, with an acceptable speed trade-off when running on GPU.

Model Results (YOLOv11m vs YOLOv11L)

Two model sizes were evaluated on the validation set.

Metric	YOLOv11m	YOLOv11L
Best mAP@0.5	0.725	0.779
Best mAP@0.5:0.95	0.390	0.429
Best Precision (class)	0.857 (alligator cracks)	0.889 (potholes)
Best Recall (class)	0.920 (alligator cracks)	0.859 (alligator cracks)
Worst class (mAP@0.5)	white line blur (0.295)	white line blur (0.295)
Inference speed	16.3 ms/image	21.9 ms/image

Note: YOLOv11L provides higher detection performance, but YOLOv11m is faster. For this application, YOLOv11L was chosen to prioritize detection quality in real road conditions.

Evaluation Plots (YOLOv11L)

Precision–Recall Curve

Confusion Matrix (Normalized)

Performance & Hardware

• Recommended: NVIDIA GPU (CUDA) for smooth real-time inference
• CPU Mode: Supported, but not suitable for real-time usage due to lower FPS and slower responsiveness

Packaging & Deployment

The application is packaged for Windows distribution using:

• PyInstaller (bundles the Python application)
• Inno Setup (creates an installation package)

Project Structure

• src/ — application source code
• src/assets/ — UI assets/resources
• src/models/ — model weights location
• screenshots/ — sample detections, UI previews, and evaluation plots

Additional Screenshots

More screenshots are available in the screenshots/ folder, including:

• Real-time detection output (screenshots/demo/demo-realtime.png)
• Video detection output (screenshots/demo/demo-video.png)
• Video upload (screenshots/demo/ui-video-upload.png)
• Report generation previews (screenshots/demo/report-excel.png, screenshots/demo/report-pdf.png)

Notes

Model weights may be excluded from the repository due to file size.