Date of Publication

6-2022

Document Type

Bachelor's Thesis

Degree Name

Bachelor of Science in Premed Physics

College

College of Science

Department/Unit

Physics

Honor/Award

Outstanding Thesis

Thesis Advisor

Romeric F. Pobre

Defense Panel Chair

Enrique M. Manzano

Defense Panel Member

Michelle T. Natividad
Reggie C. Pantig

Abstract/Summary

We coupled the M5Stack UnitV2 AI camera in tandem with the AMG8833 thermal camera in developing a smart imaging system in temperature monitoring and face mask image classifier in maintaining social distance between test subjects in following post pandemic in-person school campus safety protocol. Calibration of both camera systems were initially done with gold standard equipment (e.g. thermal gun and meter tape) prior to any field test to ensure accuracy and precision in the temperature and distance detection. The calibration lines of -0.08T + 38.95oC and 0.035R + 0.96 m were applied to the AMG8833 thermal camera and M5Stack UnitV2 AI camera distance respectively where T is the temperature in oC from 0 oC to 100oC and R is the range in meter from 0.5 meter to 1.5 meter only. A living human test subject was enrolled in the field test measurement at the corridor of Science and Technology Center (STRC) building of De La Salle University, Manila campus that consists of three detection targets: Temperature Detection, Distance Detection through Motion Tracking and Face Mask Detection. Field test ranges from 0.5 meter to 4.0 meters at 0.5 meter step size were done with ten (10) trials for each test range. This would yield a total of 80 trials for all field tests that consists of two (2) physical quantities and one image classifier: temperature, distance, and human with face mask. The AMG8833 thermal camera was pitted against the Optris Pi 450 thermal camera to compare its range performance with actual temperature of the human body onsite and standard heat source. Results showed that AMG8833 thermal camera has a shallower slope at (0.14 °C/m) and a higher overall accuracy of 99.49% compared to Optris Pi 450 at (0.93°C/m) and lower overall accuracy of 96.01% in terms of its distance dependence temperature performance. Although the Optris Pi 450 camera has higher thermal image resolution compared to the AMG8833 camera, correlation statistical analyses have shown that the AMG8833 thermal camera is more correlated to non-contact thermal gun than with Optris Pi 450. Thus, the AMG883 thermal camera is more suitable than Optris Pi 450 in fetching human body temperature on test range from 0.5 meter to 4.0 meters for distance dependent temperature monitoring. The study infers that the two sensors, AMG8833 and M5Stack UnitV2 can work independently with low correlation during the combined detection test for temperature and distance respectively. Additionally, the M5Stack UnitV2 AI camera showed acceptable face mask image recognition with an overall specificity of 89.80%, sensitivity of 97.56 %, precision of 88.89% and accuracy of no mask 93.33% computed from the confusion matrix for distances from 0.5 m to 1.5 m that displays “Mask Detected” and “No Mask Detected” in the monitor interface with the M5Stack platform.

Abstract Format

html

Language

English

Format

Electronic

Physical Description

192 leaves

Keywords

Cameras; Imaging systems

Upload Full Text

wf_yes

Embargo Period

11-6-2022

Share

COinS