Detection of Fruit from an Image Using a Single-Shot Detector for Accurate Calorie Estimation

Calorie intake is always a great concern for people of all ages and kinds in today’s world. Consuming more calories than required raises problems such as excessive fat storage, which leads to several fatal health issues. Consuming fewer calories also leads to a lack of proper nourishment for the body, which is just as dangerous as overconsumption. So, keeping track of calories is vital, and this needs to be done in less time to accommodate the busy lives of people. This work shows the implementation of a model that helps people take a picture of their meal and estimate the number of calories present in the fruits from the image. A Convolutional Neural Network based on Single Shot Detector is used to detect different fruits present in the image based on their color and shape features. Single Shot Detector has a higher accuracy and a lower computation time than the other traditional algorithms and produces a detection accuracy of 99%. The number of calories in different fruits is then estimated based on the area occupied by the fruit in the resized image. Though the model was initially developed for fruits, it can also be easily extended to a wide range of varieties. This work emphasizes the use of object detection models for accurate calorie estimation of fruits in images. The system showed an overall accuracy of 87.02% for calorie estimation. Finally, a mobile application is built on the techniques of fruit detection and calorie estimation to estimate the number of calories present in a diet.