Skyebase_concrete_cracks

Table of contents

• General info
• Results
• Technologies
• Content

General info

The goal of the project is to detect cracks in concrete above and under water. And the detection of crack width and length and surface area.

Results

• make opencv draw a contour aroun the cracks, getting the surface area of the crack with the contourArea function.
• segment the crack from the background based on a RGB treshold, being able to get the surface area based on RGB pixel values.
• draw a bounding box around the crack area based on the contour x&y coordinates, but could not automatise it with cv2 for all images.
• use opencv to life detect cracks based on threshold pixel values.
• detect crack objects using yolo, opencv and Darknet, drawing a boundingbox around the crack object
• trained a CNN model based on open source data from Kaggle, accuracy was only 50% (should probably used pretrained CNN)

Technologies

• CNN
• Darknet
• opencv
• yolov3
• labelimg

Content

• cracks.py -> loading in the data
• train_CNN_model.py -> training the data with CNN model
• contour.py -> drawing contour around cracks in images
• contour_area.py -> saving surface areas of cracks in df
• crack_pixel_counter.py -> saving surface areas of cracks in df
• irl_object_contours.py -> detecting cracks irl with camera
• object_detection_test.py -> training model to detect crack, draw boundingboxes around it
• google_collab_train_darknet.py -> train darkent pretrained CNN with crack images, where i manually labeled the images with 'labelimg'