Industry: Transportation and Municipal Services
Business Impact: Streamlined road maintenance, reduced repair costs, improved road safety
Company Size: Medium-sized enterprise
Project Idea:
Road defects like potholes significantly impact vehicle maintenance costs and safety. The aim was to leverage AI and computer vision to detect these road defects, allowing municipalities to automate road inspections and repair cost estimations.
Challenges:
- Data Scarcity: Lack of high-quality images of potholes hindered initial machine learning model training.
- Data Quality Issues: Variability in pothole appearance due to lighting and vehicle perspectives affected data consistency.
- Manual Data Mapping: High time investment required to manually map data from video footage.
Prototype Description:
Developed a Python script that processes images and videos to identify potholes and other defects. The system analyzes video clips frame by frame, marking potential defects for easy identification.
Results:
The proof of concept successfully detected road defects but struggled with complex classifications. The accuracy improved with data volume, indicating the need for extensive training sets.
Improvements:
To enhance classification accuracy, the integration of multiple algorithms and methods is proposed, reducing dependency on a single model and increasing efficiency in data processing.