首頁 » Multiview three-dimensional reconstruction by millimetre-wave portable camera

Multiview three-dimensional reconstruction by millimetre-wave portable camera

by flliang35@gmail.com
0 comment

Millimetre-wave imaging is a powerful non-destructive inspection technique which has become widely used in areas such as through-the-wall imaging or concealed weapon detection. Nevertheless, current systems are usually limited to either a single view point providing a limited 3D millimeter-wave model or a multiview relying on the accurate movement of a robot arm through precise positions resulting in very bulky systems. Here we present a set of techniques to achieve a multiview millimetre-wave scanner. The aperture of the scanner is kept below 16 cm so it can be portable and, consequently, multiview can be achieved by simple hand movements. In addition, optical images are also acquired with a two-fold purpose: i) building a complementary 3D-model by employing Structure from Movement (SfM) techniques; ii) estimating the scanner position and poses. The proposed technology is illustrated for people screening, proving the capacity of the system to detect hidden weapons.

Latest Progress

Properties of the electromagnetic (EM) waves along the EM spectrum (i.e., X-rays, visible light, infrared, terahertz, microwaves, etc.) are extremely heterogeneous enabling a wide variety of applications. Among these portions of the spectrum, millimetre-waves (mm-waves), covering the range from 30 GHz to 300 GHz, provide a good trade-off between penetration capabilities and wavelength size. Consequently, they are widely used for EM imaging as they can be employed to generate images with resolution similar to conventional optical images. For this reason, mm-waves have become very appealing for fields such as security or non-destructive evaluation. Furthermore, they lay in the non-ionizing area of the EM spectrum and, consequently, they are harmless for human beings.

There is a large variety of mm-wave imagers based on different strategies. For example, it is usual to classify the systems as passive or active depending on if they require emitting or not some kind of wave. Among passive systems, radiometers are able to capture the spontaneous radiation of bodies at a given temperature by using high gain antennas so that information of small spots can be collected yielding the final image. In general, these setups have been used to image in the range of tens of metres due to the distance required by the involved large antennas to focus the scene. Focal plane arrays (FPAs), comprising a set of lenses and an array of receivers, are also able to capture spontaneous radiation but their working principle is similar to a conventional photography camera. Nevertheless, the use of these imagers requires large frequency bandwidths in order to collect enough radiation for retrieving the image. Thus, they are mostly deployed at submm-wave bands where larger bandwidths are available.

Active systems, based on illuminating the object under test by a certain source and capturing the reflected power, are also widesprea. As in the case of radiometers, active systems based on high gain antennas have been demonstrated. Active imaging can also be done at ranges of a few centimetres by resorting to near focusing lenses that are mechanically moved to perform a raster scan. Finally, real-time imaging at distances in the order of a meter has also been validated by synthetic aperture radar (SAR) techniques combined with hybrid electronic/mechanic or fully electronic approaches.

Mm-wave technology has been boosted in the last years due to these imaging applications as well as others such as high capacity radiolinks or automotive radar. Consequently, a wide variety of Monolithic Microwave Integrated Circuits (MMIC) is available reducing the cost and easing the implementation of mm-wave setups. These advances in mm-wave technology have enabled the implementation of the first generation of portable mm-wave scanners and fully electronic cameras.

Independently of the imaging strategy, EM scanners usually consider data acquisition from a single point of view. Although the images can provide an astonishing degree of quality, even containing information relative to depth, this kind of acquisition is limited by its own nature. For example, when scanning a human being, information from the backward view is completely occluded forcing the use of a second scanning panel behind the subject under test. In short-range imaging, active mm-wave scanners, which usually comprise transmitters and receivers placed very close to each other (i.e., a monostatic or quasimonostatic setup), suffer from problems when the target is not illuminated by a wave normal to its surfaces. Since the impinging waves are bounced into a direction which is different from the transmitting/receiving direction, only a small fraction of the reflected energy is received. This problem is alleviated by multistatic scanners, with dissociated transmitters and receivers positions, enabling a wider observation range. Other alternatives, including commercial systems, employ cylindrical scanning based on the accurate movement of a robot arm.

Last generation of scanners tries to bypass the aforementioned difficulties. For example, a walk-through imaging system has been presented so that it can consider snapshots at different positions in order that multiple views are available. However, each view is independently considered and not merged into a global model. Each view can be independently considered or merged into a global model. Other authors have focused on reducing the complexity of the required electronic by resorting to metasurface antennas with capacity to produce a wide set of radiation patterns by a simple frequency sweep.

On other side, conventional optical cameras have been able to benefit from multiview acquisitions for many years, enabling the possibility of building three-dimensional (3D) models or photogrammetry. This kind of techniques can be implemented either by a multicamera-setup, where the positions are accurately known, or by Structure from Motion (SfM) techniques where the camera (or, alternatively the object) is arbitrarily moved and, consequently, the position is also estimated from the images themself. Furthermore, it has been demonstrated in the recent years that the algorithms supporting this technique can run on real-time even on (relatively) low performance devices such as smartphones.

The research demonstrate the possibility of multiview with a mm-wave camera for short-range imaging. Thus, 3D images benefiting from the penetration capabilities of the EM waves can be calculated. Furthermore, working frequency and bandwidth match those of available commercial solutions which exploit relatively inexpensive MMIC for automotive radar to build the RF system. It is also relevant to mention that the positioning of the camera is based on the information extracted from a conventional camera by resorting to SfM. Thus, a 3D conventional model is also simultaneously built resulting in a more insightful inspection system.

You may also like

Leave a Comment