Depth Comparison of Objects in 2D Images Using Mask RCNN

Title

Depth Comparison of Objects in 2D Images Using Mask RCNN

Creator

Singh H.; Kirubanand V.B.

Description

Getting distance of an object from a single 2D image has always been a task. Due to various reasons, it was difficult to compare from images whether an object is closer or farther from camera. In this paper, we propose an idea to compare multiple images taken from same focal length cameras and specifying the distance of an object in those images with respect to each other. Our dataset contains images of palm of hand with particular distance from camera, and the output difference can specify in which image the palm is closer to camera as compared to others and vice versa. For this model, we are using Mask RCNN to recognize the object; in our case, it has been trained to identify palm, and then giving the output of masked RCNN to a depth identifier model to specify the distance of the palm from the camera. Directly using depth identifier model cannot give correct output as distance of background from camera results in different value for distance of targeted object in different images. So, we will be using mask RCNN to specify which part of image depth model should find distance from the camera. In the final step, we take the output of the depth model and take the mean of the output generated by it and compare the means of various images to specify relative distance from each other. 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

Source

Lecture Notes in Electrical Engineering, Vol-806, pp. 81-88.

Date

2022-01-01

Publisher

Springer Science and Business Media Deutschland GmbH

Subject

Depth analysis of object in image; Masked RCNN; Relative distance of object in images

Coverage

Singh H., CHRIST (Deemed to be University), Bangalore, India; Kirubanand V.B., CHRIST (Deemed to be University), Bangalore, India

Rights

Restricted Access

Relation

ISSN: 18761100; ISBN: 978-981166447-2

Format

Online

Language

English

Type

Conference paper

Identifier

https://doi.org/10.1007/978-981-16-6448-9_10

https://www.scopus.com/inward/record.uri?eid=2-s2.0-85122508968&doi=10.1007%2f978-981-16-6448-9_10&partnerID=40&md5=a8b5a2a97015931da5f7eb6c6cde72e6