by Nguyen, M, Tran, H, Le, H and Yeap, W
Abstract:
This article describes the design and process of a 3D Mini Camera Lens to be utilised in multi-purpose 3D stereo reconstruction on computers, mobile devices and robots. It can also be useful for various tasks including demonstration, knowledge-sharing, online programming, database sharing, results comparison in the stereo vision research community, and recreation used by non-specialists. In contrast with many existing systems that use stereo cameras, our proposed one only use a single camera. However, the acquired image streams are divided into two side by side images by a 3D Mini Camera Lens. The use of the lens has some advantages over the state of the art ones. It is very low-cost, at 5-7 USD; it can be implemented with any small image sensors of web cams, smartphones, tablets. The lens is light weight, simple to assemble and disassemble, convenient and portable; and its two stereo pictures are perfectly synchronised (as the same image sensor produces them). We develop a fully functioning system which horizontally aligns images and calculates the dense distance map of the scene using a guided semi-global stereo matching algorithm. The simple horizontal and vertical image alignment is very lightweight, using adaptive methods, and is carried out during the stereo matching; thus, no separated processes are needed. The semi-global stereo matching uses full or partial profiles generated by local block matching algorithms to achieve an efficient dynamic programming stereo matching and produce higher-quality disparity maps. The system is, therefore, plug-and-play, no special setup or calibration is required. The system gradually estimates the best parameters, including disparity range, to produce the most suitable depth image. The evaluated results have shown that this system was demonstrably accurate, fast, and relatively robust for stereo correspondence. It could be used in many simple stereo vision system such as computer games, gesture control, robotic vision, self-flying drones or -driving cars.
Reference:
Adaptive stereo vision system using portable low-cost 3D mini camera lens (Nguyen, M, Tran, H, Le, H and Yeap, W), IEEE, 2017.
Bibtex Entry:
@misc{nguyen2017adaptivelens,
author = {Nguyen, M and Tran, H and Le, H and Yeap, W},
organization = {Auckland},
publisher = {IEEE},
title = {Adaptive stereo vision system using portable low-cost 3D mini camera lens},
year = {2017},
abstract = {This article describes the design and process of a 3D Mini Camera Lens to be utilised in multi-purpose 3D stereo reconstruction on computers, mobile devices and robots. It can also be useful for various tasks including demonstration, knowledge-sharing, online programming, database sharing, results comparison in the stereo vision research community, and recreation used by non-specialists. In contrast with many existing systems that use stereo cameras, our proposed one only use a single camera. However, the acquired image streams are divided into two side by side images by a 3D Mini Camera Lens. The use of the lens has some advantages over the state of the art ones. It is very low-cost, at 5-7 USD; it can be implemented with any small image sensors of web cams, smartphones, tablets. The lens is light weight, simple to assemble and disassemble, convenient and portable; and its two stereo pictures are perfectly synchronised (as the same image sensor produces them).
We develop a fully functioning system which horizontally aligns images and calculates the dense distance map of the scene using a guided semi-global stereo matching algorithm. The simple horizontal and vertical image alignment is very lightweight, using adaptive methods, and is carried out during the stereo matching; thus, no separated processes are needed. The semi-global stereo matching uses full or partial profiles generated by local block matching algorithms to achieve an efficient dynamic programming stereo matching and produce higher-quality disparity maps.
The system is, therefore, plug-and-play, no special setup or calibration is required. The system gradually estimates the best parameters, including disparity range, to produce the most suitable depth image. The evaluated results have shown that this system was demonstrably accurate, fast, and relatively robust for stereo correspondence. It could be used in many simple stereo vision system such as computer games, gesture control, robotic vision, self-flying drones or -driving cars.},
doi = {10.1109/M2VIP.2017.8211465},
startyear = {2017},
startmonth = {Nov},
startday = {21},
finishyear = {2017},
finishmonth = {Nov},
finishday = {23},
conference = {24th International Conference on Mechatronics and Machine Vision in Practice (M2VIP)},
}