Automated Classification of Test Tubes Based on Uncontrolled Image Analysis

Metrology, the science of measurement, involves defining and establishing standards for quantifying physical quantities. A critical aspect of this discipline is to accurately measure and understand the relationships between object dimensions. However, determining dimensions from images poses several challenges, including optical distortions, lighting errors, and perspective issues, which can distort the perceived dimensions of objects. Additionally, proper calibration of images and knowledge of exact real-world reference dimensions are essential for achieving precise and accurate measurements. In this paper, we propose a solution, based on a Convolutional Neural Networks (CNNs), for classifying three different types of test tubes from images. The test tubes exhibit slight dimensional variations (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{16}\mathbf{mm}\times \mathbf{100}\mathbf{mm}, \mathbf{13}\mathbf{mm}\times \mathbf{100}\mathbf{mm}$</tex>, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{13}\mathbf{mm}\times \mathbf{75}\mathbf{mm}$</tex>). The images are captured in uncontrolled conditions, without specific guidelines for the operator, and lack reference dimensions to aid in identifying the test tube type. To address the absence of reference dimensions, our approach relies on capturing and analyzing subtle visual cues and patterns unique to each test tube type. By training on a diverse set of samples, our model can generalize effectively to unseen images, providing robust and accurate classification results. Experimental evaluations demonstrate the effectiveness of our proposed method, achieving high classification accuracy even under challenging conditions with varying lighting, backgrounds, and orientations. The results underscore the potential of our approach in practical applications where manual classification based on visual inspection alone is unreliable or time-consuming.