Real-Time Detection and Clasification System of Biosecurity Elements Using Haar Cascade Classifier with Open Source

This document presents an image processing for the detection and classification of biosecurity elements in real time by means of cascade classifiers. The operation was based on a Haar Cascade Classifier and data augmentation to complete the datasets. The images were acquired using an embedded Raspberry Pi 3B+ system connected to a Raspberry camera and then processed in Python. Both OpenCV and the Cascade Trainer GUI application, available for Windows versions 7 or higher, were used to create the classification models, so the images captured by Raspberry Pi had to be transferred to a personal computer. There were 4250 images that were converted by data augmentation techniques to 25401, with an average data increase accuracy of 88.492%. Also, 5 classification models were obtained corresponding to 5 categories of biosecurity elements referring to mask, gloves, glasses, anti-fluid clothing and anti-fluid footwears, with success rates in the classification of 90.2%, 92.7%, 92%, 89.7% and 94.1% respectively. In addition to the performance tests according to the hit rates, the system was evaluated by measuring the processing response time, obtaining fluctuating times between 0.475 seconds and 0.571 seconds.