Identify and Counting Number of Objects of Image Using TensorFlow Lite in a Mobile Application

Yap, Eng Seong (2023) Identify and Counting Number of Objects of Image Using TensorFlow Lite in a Mobile Application. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text RSD_Yap Eng Seong_Fulltext.pdf Restricted to Registered users only Download (4MB)

Abstract

This paper talks about a research study of object counting, which is a process of detecting objects from an image and counting the number of each object detected. Object detection is the main component of object counting, which includes the processes of image localization and image classification to write out the location of objects in an image followed by determining the types of objects present in the image. The aim of this project is to solve the problem of saving time to count a moderately high number of objects that are piled together. An object counting system is created in this project to count the retail store products on the shelves of retail stores or supermarkets. Android Studio is used to develop the system in Android platform while TensorFlow Lite (TFLite) is used as the object detection component of the object counting system which is compatible with Android devices. Photos of retail store products are also taken from a real physical store to use as the dataset for the TFLite model for object detection. Incremental model is used in this project, while the requirements of the system are collected from various research and journal articles that talk about object counting, object tracking and retail store products management. Unit testing is done to the object counting system to ensure all the modules work properly to perform the object counting process. Overall, this project presents some of the research on object detection and object counting, while the object counting system created is targeted to help in the tracking of retail store products in the store. The system is not physically implemented in real retail stores to be used as the system is very simple and does not work exactly as intended.