A YoloX-Based Approach for Detecting Human and Classifying Their Own Pet to Monitor and Improve Services at the Veterinary Clinic

Nowadays, the YoloX-based approach is innovative in the field of computer vision. In this paper, a model based on it will be designed, trained and tested. A dataset with thousands of image of humans and their pets are collected, selected, labelled and annotated manually. Next, the YoloX-based model is trained on this dataset of 1118 images in total in order to recognize human with pet and classify different pets such as dogs and cats. The objects are trained independently so that the model can recognize the objects “Human”, “Dog”, “Cat” most accurately. After training the model, a proposed algorithm in this paper is applied to find the coordinates of the objects and display them. Finally, the trained model is implemented in a monitoring system for testing. The system was then tested experimentally in the field for recognition accuracy according to mAP@.5 measurements as follows: Cat recognition: 99.8%, Dog recognition: 90.8% and Human recognition: 98.3%. The system accuracy reaches to an average of 96.31%.