Autonomous Concrete Crack Detection and 3D Analysis Using Husky AGV
Disciplines
Robotics
Abstract (300 words maximum)
Several studies have been conducted in the field of automated concrete crack detection using different analysis methods and robotic vehicles; however, successful 3D crack analysis using a movable camera mount remains to be accomplished. Our research aims to streamline the current concrete crack analysis process to determine the severity of the crack using manual crack length and depth inspection. For this research, a RealSense depth camera will be attached to the robot arm on the front of a Husky AGV to inspect a crack in concrete from multiple angles. A LiDAR unit, using a YOLO machine learning model, will also assist an operator in finding cracks. The additional scans from the different camera angles will be used to create a depth model of the crack using a cloud point grid, which will be computed on a Raspberry Pi 4. The success of the LiDAR and machine learning model will be measured using the percentage of positive results, false positive results, and negative results compared to the total amount of detections. The camera scans will be compared to manually collected results and examined for similar measurements.
Academic department under which the project should be listed
SPCEET - Robotics and Mechatronics Engineering
Primary Investigator (PI) Name
Muhammad Hassan Tanveer
Autonomous Concrete Crack Detection and 3D Analysis Using Husky AGV
Several studies have been conducted in the field of automated concrete crack detection using different analysis methods and robotic vehicles; however, successful 3D crack analysis using a movable camera mount remains to be accomplished. Our research aims to streamline the current concrete crack analysis process to determine the severity of the crack using manual crack length and depth inspection. For this research, a RealSense depth camera will be attached to the robot arm on the front of a Husky AGV to inspect a crack in concrete from multiple angles. A LiDAR unit, using a YOLO machine learning model, will also assist an operator in finding cracks. The additional scans from the different camera angles will be used to create a depth model of the crack using a cloud point grid, which will be computed on a Raspberry Pi 4. The success of the LiDAR and machine learning model will be measured using the percentage of positive results, false positive results, and negative results compared to the total amount of detections. The camera scans will be compared to manually collected results and examined for similar measurements.