Impact of Camera Hardware on the Performance of Convolutional Neural Network Models in Machine Vision Applications

Session

Mechatronics, Sysem Engineering and Robotics

Description

In the rapidly growing field of artificial intelligence (AI), machine vision is an important area with applications ranging from agriculture to healthcare and improving people's quality of life. A critical factor in the effectiveness of AI models, especially in machine vision, is the complicated interplay between hardware and software parameters. This study addresses the performance metrics of a Convolutional Neural Network (CNN) model, focusing on the influence of different camera hardware. Since the CNN model used in this case is tailored for regression-based predictions, its evaluation depends on the number of predictions made over a period of time. Preliminary results highlight that camera hardware attributes can increase the prediction rate by achieving more than 10 predictions within a 15-second window, with this rate escalating over longer durations. These findings clearly indicate that the performance of the CNN model can be improved simply by selecting an appropriate camera model, without requiring changes to the CNN parameters or training datasets.

Keywords:

Convolutional Neural Network, Machine Vision, Regression

Proceedings Editor

Edmond Hajrizi

ISBN

978-9951-550-95-6

Location

UBT Lipjan, Kosovo

Start Date

28-10-2023 8:00 AM

End Date

29-10-2023 6:00 PM

DOI

10.33107/ubt-ic.2023.89

This document is currently not available here.

Share

COinS
 
Oct 28th, 8:00 AM Oct 29th, 6:00 PM

Impact of Camera Hardware on the Performance of Convolutional Neural Network Models in Machine Vision Applications

UBT Lipjan, Kosovo

In the rapidly growing field of artificial intelligence (AI), machine vision is an important area with applications ranging from agriculture to healthcare and improving people's quality of life. A critical factor in the effectiveness of AI models, especially in machine vision, is the complicated interplay between hardware and software parameters. This study addresses the performance metrics of a Convolutional Neural Network (CNN) model, focusing on the influence of different camera hardware. Since the CNN model used in this case is tailored for regression-based predictions, its evaluation depends on the number of predictions made over a period of time. Preliminary results highlight that camera hardware attributes can increase the prediction rate by achieving more than 10 predictions within a 15-second window, with this rate escalating over longer durations. These findings clearly indicate that the performance of the CNN model can be improved simply by selecting an appropriate camera model, without requiring changes to the CNN parameters or training datasets.