Coherent Vision Can Help Machines See Like Humans – EE Times

The dream of autonomous robots that can help with chores, chauffeur us around and replace factory workers has been around for longer than the Jetsons series aired in the 1960s. The reality is that many companies have promised the availability of self-driving cars by 2020, but we’re all still waiting. And the situation is not too different in the service robotics and industrial automation sector.
Why is it taking so long?
One key aspect is how machines perceive their environment. While robotic vision has come a long way in the last decade, in part enabled by higher resolution cameras and new 3D vision technologies, it still is far less efficient than humans in perceiving its environment.
In fact, the current approaches might have gotten it all wrong from the start. For example, we humans don’t digitize our surroundings in multi-megapixel resolution and then meticulously try to find contours of objects and compare those frame by frame to conclude their motion paths.
Imagine us sifting through a multi-megapixel image with a magnification glass, marking up slight variations in contrast and color as possible objects, and then doing this at ten frames per second to track and predict the motion of hypothetical objects.
Human perception works differently. Our eye is an extension of our brain, preprocessing vast amounts of information. As a result, we have many more motion-sensitive cells in our retina, making us immediately aware if something moves in our field of vision. Only then do we use our high-resolution cells in the macula to identify the objects of interest—a far more efficient way of object recognition, tracking and prediction.
Enter a new 3D vision technology available now to help machines perceive their environment much more like humans. It’s called coherent vision.
The technology sends out coherent laser light and captures much more than just intensity information from returning photons. It also captures tiny frequency shifts induced by object movements and provides information about the material and surfaces sensed via polarization changes.
Existing 3D sensing technologies include direct and indirect time-of-flight approaches, projected IR patterns (a.k.a. as structured light) and triangulation techniques such as stereo vision. Not all of them offer instantaneous motion information and typically suffer from significant trade-offs in terms of range, eye safety, crosstalk immunity and precision.
Frequency modulated continuous wave (FMCW) or coherent 3D sensing
Rather than relying on detecting variation in light intensity, a common method of existing 3D sensing techniques, the coherent 3D sensing approach depends on low-power frequency chirps from a highly coherent laser. This is also known as frequency modulated continuous wave (FMCW) technique, already used in state-of-the-art radar sensors.
Coherent photons can travel hundreds of meters, interact and pick up features of the target, then return while remaining in a coherent state where they can be mixed with a portion of the outgoing light for near lossless amplification.
Mixing returning with outgoing photons results in a beat frequency that is down-converted from the optical frequency, that is in the terahertz region, to the low gigahertz region and can be easily analyzed by available electronic circuits.
The distance of the measurement is reflected in the form of an optical frequency shift. If the measurement point also has a radial velocity, the reflected chirp adds a doppler frequency shift.
Using an up-and-down chirp allows coherent 3D sensors to instantaneously resolve both the range and velocity of each pixel. This capability is effectively expanding the 3D sensing to 4D, meaning sensing x, y, z, and velocity vectors of an object at the same time.
The mixing of the returning photons with a portion of the outgoing laser light results in a nearly lossless optical amplification, allowing a much higher detection sensitivity and accuracy.  Due to the higher detection sensitivity of coherent systems, a laser power level in the hundred milliWatt region is typically sufficient to measure objects hundreds of meters away, allowing this technology to be integrated onto a chip for use in mobile applications.
Linearly polarized photons additionally can change their polarization state when interacting with targets, allowing the detection of material and surface characteristics, such as windows or human skin.
FMCW advantages and 4D machine vision
Technology advances and reduced cost have made 3D vision a critical technology in industrial manufacturing automation used to enhance productivity, efficiency and quality. With many competing technologies available, the choice of technology is often decided by the applications, ranging from general quality inspections, validations, verifications and sorting along with safety and security.
FMCW promises to improve performance vectors in several dimensions, allowing for higher precision scanning at longer ranges while at the same time being eye-safe and immune to outdoor lighting conditions or multi-system crosstalk. On top of that, it offers native 4D vision by providing velocity information with every measurement.
What’s the holdup?
Why has it taken so long to make coherent 3D sensing systems mainstream?
The key challenge in creating an FMCW solution has been the low-cost, high-volume manufacturing of high-performance components. The coherent approach requires lasers with long coherence lengths (narrow linewidths) and coherent light processing to extract additional information carried by photons.
This requires very accurate and low-noise optical signal processing circuits to form a coherent receiver. In addition, polarization plays a role here, as coherent beating will only work for photons of the same polarization. The laser source’s wavelength stability and linearity are critical during measurement; otherwise, the signal-to-noise ratio can be degraded significantly.
Creating such a stable, robust, and accurately defined optical system with discrete components is very challenging and expensive. In order to address this, SiLC Technologies has created a solution that integrates all needed optical functionalities into a single silicon chip using semiconductor manufacturing processes used to manufacture electronic ICs.
In other words, the same approach behind very complex electronic circuits integrated into silicon that already enable consumer products at very low cost, can now be deployed to make highly complex optical circuits for photonics applications.
The silicon photonics integration platform integrates high-performance components into a single chip using mature semiconductor fabrication processes, offering a low-cost, compact, and low-power solution. Silicon manufacturing also offers affordable, high-volume scaling of complex devices and technologies.
In summary, 3D vision is critical for machine perception. Coherent 3D sensing using the FMCW technique is the newest of these technologies, expanding the performance characteristics of vision systems on many levels–even into the fourth dimension. Rather than relying on the time of flight, stereo vision, triangulation or structured light, FMCW sensing takes advantage of the properties of photons themselves.
Leveraging this approach has been prevented by the cost and the required number of components. The power of silicon integration and the heritage of bringing silicon photonics products to market can be harnessed to at last commercialize a cost-effective coherent vision sensor. Using additional instantaneous velocity information, this will help machines to perceive their environment more like humans.
– Ralf J. Muenster is Vice President of business development and marketing for SiLC Technologies, Inc.
You must Register or Login to post a comment.
This site uses Akismet to reduce spam. Learn how your comment data is processed.