Computer Vision: The Future of Image Analysis for Paranormal Research?
Spoiler Alert: The image on display is not of a bespectacled ghost, but rather it is of a kinetic sculpture of a bicyclist in motion, and rendered from a computer vision algorithm designed for change detection and movement tracking
Computer Vision is the science of how a computer recognizes, interprets and understands the environment it sees. The four approaches of computer vision can be broadly characterized as: recognition; reconstruction; registration; and reorganization
Computer vision is the power behind optical character recognition; 3-D and panoramic viewing; face and object recognition; biometrics and medical imaging; robotics; interactive games; special effects in motion pictures and television; smart as well self-driving cars; and helping the blind to navigate
Many computer vision applications can be found in the OpenCV (Open Source Computer Vision) library which provides a cross-platform repository containing more than 2500 algorithms for machine perception.
OpenCV algorithms that can be used to identify or track moving objects and to recognize faces or features in scenery
may have usefulness for paranormal research. Some examples classes of algorithms include:
- Background Subtraction. Mixture of Gaussians (MOG) is a filtering technique that extracts a moving foreground from a static background, which is useful for change detection. The MOG2 variant used in the feature image also tracks shadows in connection with moving objects
- Optical Flow. These are methods for object tracking arising from either the movement of the object or the camera. The Lucas-Kanade-Tomasi (L-K-T) algorithm uses corner points to track object motion. Gunner-Farneback is an algorithm for dense optical flow that uses colors and intensity for object direction and speed
- Edges and Gradients. The Canny algorithm is a multi-stage noise filtering method for edge detection. Laplace and Sobel are noise filtering algorithms designed for detecting image gradients (or changes in image intensity)
- Image Thresholding. These algorithms are designed to see in different lighting conditions. The adaptive threshold Gaussian technique as an example auto adjusts to lighting conditions
- Face/Object Recognition. These methods makes use of trained models (classifiers) that
in a staged (cascading) manner
look for line, edge and rectangular features in pixel blocks that correspond to facial features. The Haar cascade is commonly used and it is owned by Intel Corp
Perhaps the easiest way to use computer vision scripts is within Python, which is a high-level language general purpose language designed with a simple and readable syntax. OpenCV scripts can also be customized and compiled in C++, MATLAB and CUDA.
CUDA runs on NVIDIA graphics processing units (GPUs) providing acceleration that can enable computer vision in real time. ANACONDA is a distribution bundles CUDA, Python and its most popular packages.
Technologists still do not fully understand how computers see. An example would be apparent false positives in the form of person-like articulated skeleton anomalies occasionally encountered in the Kinect structured light sensor
We can not claim computer vision algorithms will find ghosts nor see what is unseen. However, computer vision carries to the potential to normalize image analysis, putting it on par with industry practices elsewhere. It will prove useful to believers and skeptics alike
Background Foreground Substraction MOG2 Algorithm Applied to Bicycle Mini Kinetic Sculpture from Dana Paige. (2018). Maryland Paranormal Research
. All rights reserved.
Hays, J. (2018). CS-143: An Introduction to Computer Vision. Brown University, Providence RI
How Computer Vision Is Finally Taking Off, After 50 Years. (2017, May 3). Nat and Friends. YouTube
OpenCV: Open Source Computer Vision Library. (2018). The OpenCV Team
Pulli, K. et al. (2012, Jun). Real Time Computer Vision with Open CV. Communications of the ACM. Association for Computing Machinery. Reprinted by NVIDIA Research