Object Detection in Refrigerator and Calorie Estimation using EfficientDet1 and YOLOv5S Techniques on Mobile Devices
Date
2023-04-28
Authors
Agarwal, Rohit
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
In this research, we developed a tool to automate object detection and calorie estimation
of food items in home refrigerators compatible with both windows and mobile
devices. This tool detects objects using the state-of-the-art one-stage methods EfficientDet1
and Yolov5s (You look only once). The tool’s performance is assessed by
comparing speed, accuracy, time, and mean average precision (mAP) to learn the
advantages in real-time applications. Open-source frameworks such as TensorFlow
or PyTorch are used to train the object detection models. Using these models, the
object detection of items is identified and classified. A Python algorithm is developed
to count the number of items and their calorie estimations. The algorithm automates
the process by generating the inventory of items in the refrigerator and sending it to
users at their designated e-mail addresses. If any of the desired food items are not
present or fall below their threshold values, then the items are flagged.
The experimental results indicate that object detection models (EfficientDet1 and
Yolov5s) are suitable to run on smaller models for real-time applications. The associated
web-based graphical user interface (GUI) (compatible with mobile devices
as well) displays the list of items in a detected image, the calorie estimation of each
item, and the list of missing items needed for further action. Though our GUI Python
code is tested on the local computer, it can be repurposed on any platform such as
mobile or embedded devices. The advantage of automatic detection of refrigerator
items through the mobile app is helpful in making life easier and healthier, especially
for diet-conscious people.
Description
Keywords
Object Detection