Determining Image Scale in Real-World Units Using Natural Objects Present in Image

EasyChair Preprint 9589

Abstract

The image is a 2D representation of intensity values in rows and columns that are stored in a digital computer. Some of the images have a real-world unit scale (e.g., mm, cm) that helps individuals to assess the size of the objects present in the image. This scale is introduced at the time of acquisition of the image by the hardware system. Unfortunately, the scale is not present in most of the images, and it poses a problem to determine the actual size of the objects captured in an image. Therefore, the determination of the size of objects in an image became the main goal of this research proposal. The fact that this area of study is yet to be explored to its fullest was the motivation behind the work.

In this work, we propose to find image scale (size in real-world unit per pixel) using the common size of the objects present in the image. These natural objects could be people, cars, bikes, signposts etc., depending upon the location, traffic and time when the image was taken. The dataset which is used for the research is acquired from KITTI and the RGB-D images of the dataset have been used. It is then wielded in the derivation of the mathematical function to relate the depth of the object with the object in real-world unit per pixel. In this way, the desired outcome is achieved.

Keyphrases: Artificial Learning, computer vision, image processing, image scale, machine learning, size estimation

@booklet{EasyChair:9589,
  author    = {Saurabh Singh and Rhea S Shrivastava},
  title     = {Determining Image Scale in Real-World Units Using Natural Objects Present in Image},
  howpublished = {EasyChair Preprint 9589},
  year      = {EasyChair, 2023}}