Image Processing

Classical image processing experiments with OpenCV: vibration detection from camera images, window-based floor counting, and a visual-words feature pipeline for machine fault diagnosis.

Overview

The scripts apply traditional computer vision (filtering, frequency analysis, wavelets, keypoint detection, contour analysis and clustering) to engineering problems. Two reference papers on image-based vibration analysis and machine fault diagnosis are included as background.

Contents

Path	Description
vibration_detection_and_analysis_with_image_processing/main.py	High-pass filtering, histogram equalisation, FFT magnitude spectrum, continuous wavelet transform and SIFT keypoints; compares a test image against a reference image with SSIM, PSNR and MSE and flags excessive vibration against thresholds
predicting_floor_number_according_to_window_detection.py	Google Colab script: Canny edges, morphological closing and contour filtering detect window-like rectangles, then DBSCAN on their y positions estimates the number of floors
classification.py	Draft visual-words pipeline: integral image, SURF descriptors, k-means vocabulary and histogram feature vector (no trained classifier yet, and it relies on a load_image helper defined in main.py)
*.pdf	Reference papers: "An Image Processing Approach to Machine Fault Diagnosis Based on Visual Words Representation" and "Computer Vision Tracking Techniques Applied to Vibration Analysis"

Tech stack

• Python, OpenCV, NumPy, SciPy, scikit-image, scikit-learn, Matplotlib
• Google Colab (floor counting script)

How to run

Vibration analysis expects test_image.png and reference_image.png in the working directory:

pip install opencv-python numpy scipy scikit-image matplotlib
cd vibration_detection_and_analysis_with_image_processing
python main.py

scipy.signal.cwt and ricker were removed in SciPy 1.15, so use an earlier SciPy version. The floor counting script uses google.colab for file upload and display and is meant to be run in a Colab notebook. SURF in classification.py requires an opencv-contrib-python build with non-free modules enabled.

Author

Goktug Can Simay: GitHub | Website