DEVELOPMENT OF LONGITUDINAL AND TRANSVERSE CRACK DETECTION FEATURES IN AN AUTOMATIC ROAD PAVEMENT CONDITION SURVEY SYSTEM BASED ON VIDEO ANALYTICS

• Main Author: Jauhar Sidqi, Akmal
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/85279
• Institution: Institut Teknologi Bandung

Roadway is a vital asset for economic growth in Indonesia. To maintain the functionality of roadways, authorities need to perform maintenance on the road assets they manage. One of the efforts is by conducting a pavement condition survey. To ensure the survey process runs efficiently, previous research has developed a system to automatically and efficiently survey road conditions, named the Pavement Condition Survey System (PCSS). The current PCSS accepts input in the form of road segment videos and analyzes them using machine learning models. The existing PCSS can detect two types of road damage: potholes and alligator cracks. The objective of this research is to enhance the capabilities of PCSS by increasing the number of damage types it can detect from the previous two types (potholes and alligator cracks) to four types (adding longitudinal and transverse cracks) and automating the road damage counting process. The author adopts the Crisp DM system development methodology, a methodology consisting of six main stages: business understanding, data understanding, data preparation, modeling, evaluation, and deployment, designed to provide a structured guide in the development process. The results show that the PCSS successfully met all research objectives by generating requirements that align with user needs. The system can detect four types of road damage: potholes (precision = 1; recall = 0.91; F1-score = 0.95; and FNR = 0.09), alligator cracks (precision = 1; recall = 0.78; F1-score = 0.87; and FNR = 0.22), longitudinal cracks (precision = 1; recall = 0.77; F1-score = 0.87; and FNR = 0.23), and transverse cracks (precision = 1; recall = 0.47; F1- score = 0.60; and FNR = 0.57). It can track detected objects and output results in the form of an interactive map, a CSV file with damage information, and a detection video.