Image filtering using object detection

Akshat Vishal Sud

Bakalářská práce

Abstract:

The primary aspect of the project is associated with the detection of objects within the image by incorporating the COCO dataset. The purpose is to filter a set of user-provided images based on specified target object names. Only the images that contain said objects are a part of the resulting output bucket and can then be further used by the users. The methodology that has been used for the completion of this project includes the use of HTML and JavaScript based ML5 library as the frameworks for the program code that has been written in this study. This project utilizes a pretrained COCO SSD model for object detection. The accuracy level of the program that has been implemented has been tested using a database of new images. Based on this database, it has been identified that the program is quite efficient in identifying the different objects as per the search conducted by the user. The overall accuracy of the pretrained COCO SSD model was 86.2%. Object of interest were not properly detected in challenging scene conditions or when occluded or captured only partially.

Keywords

COCO dataset, Image processing, JavaScript

Jazyk práce: angličtina

Datum vytvoření / odevzdání či podání práce: 1. 12. 2023

Obhajoba závěrečné práce

• Obhajoba proběhla 8. 2. 2024
• Vedoucí: Ing. Petr Hanzlík, Ph.D.
• Oponent: Petr Vlachynský, Ing.

Česká zemědělská univerzita v Praze

Provozně ekonomická fakulta

Bakalářský studijní program:
Informatics