Abstract

Some diagnostic tools rely on human eyesight to determine their results. Ideally, in these cases, human eyesight needs to be verified with other people, or even technology. Misinterpretations from the human eye may lead to consequences, and it may possibly include fatal repercussions when a patient is delivered a misdiagnosis. For example, diagnoses that exhibit physical change, such as MRI scans, or ones as simple as pH strips rely on human eyesight to discover changes and contrasts from a library/database. Not only are there multiple known libraries or databases about a certain diagnostic tool, but also human eyesight is not completely reliable to determine precise results. Hence, to eliminate this issue, computer vision is used as a powerful tool to objectively “diagnose” and provide a more accurate result when viewing diagnostic tools. Through certain physical changes, such as color (BGR), computer vision can detect and differentiate with a library/database and provide more accurate results, and is customizable for an array of diagnostic applications.

Software

Raspberry Pi 4, Open CV and Python as the programming language

Hardware

The hardware consists of two detachable parts: the platform and camera mount, along with a user interface. 2 pH strips can be mounted on the platform and placed directly under the camera which is connected to the Raspberry Pi for testing. Users can interact with the touchscreen mounted above the Raspberry Pi to test the pH strips.

Software: Image Analysis without Camera

Through image analysis, it differentiates between several shades of color using BGR. To create a library, an original image is uploaded and masked to find the contours which differentiates it from its background using grey thresholding. After masking and contouring, OpenCV detects which contour is included in the library through area boundaries, and it finds a total of 64 valid contours. Therefore, the mean or average color of each contour is taken from the middle points, marked in red.

64 contours are then given an array depending on the type of indicator. It would be classified into 16 sub-arrays, where each sub-array will have another 4 sub-arrays.