Seeing Isn’t Believing – Camera and Infra-Red based TOF 3D vision systems

Infrared (IR) based light curtains have been the accepted way of protecting passengers from closing elevator doors since the 1990’s, however, over the last 5 years there have been numerous show stands exhibiting new and exciting technology, purporting to render light curtain technology obsolete. More recently the update to the North American Elevator Safety Code (ASME 17.1:2019) requires detection not just for the area between the closing doors, but also detection of approaching passengers on the landing.  Despite these bold announcements in the market little appears to have changed, why?  In this article we look to separate the facts from the hyperbole and in doing so set realistic expectations on what to expect over the coming years.

Current IR light curtain technology is conceptually relatively simple.  IR light waves are pulsed through a series of diodes running the height of the door.  At the other side of the door opening IR detectors running the height of the door look to detect these IR light waves.

A break in the signal means there is something blocking the passage of these IR light waves.  This system generates its own signal and in doing so is in full control of the frequencies and timing used.  This fact, combined with object presence detection by signal blockage rather than reflection, make IR light curtains inherently reliable.  However, IR light curtains have their limitations.  Firstly, their size and shape make them difficult to ship and store.  Secondly, and more pertinent now than ever with the launch of the ASME 17.1:2019 code, they do not normally provide any sensing of people on the landing, commonly known as 3D detection in the elevator industry.  Some light curtains provide this feature, but it relies on emitting IR light waves and detecting any resulting IR light waves that have bounced off objects (e.g. an approaching person) on the landing.  Sadly, this 3D solution has always had a couple of inherent issues: it only detects objects in a relatively small area, it can’t distinguish between moving and stationary objects and worst of all it requires the IR light to be reflected and so any objects that absorbs light (e.g. woolen clothing) are not detected.

This is very significant as it means any 3D detection system that relies on the reflection of light waves will struggle to meet this code or detect a person wearing highly light absorptive clothing. 

Whilst IR Time of Flight (TOF; a method for measuring the distance between a sensor and an object, based on the time difference between the emission of a signal (e.g. Infra-red) and its return to the sensor, after being reflected by an object) is far more advanced in the features it can provide versus a 3D light curtain, ultimately it falls foul to the same issues, namely that if the IR light waves it emits are absorbed rather than reflected, it is almost impossible to give a reliable detection signal. Furthermore, for hopes of using this technology for 2D detection, a variety of limitations in the ability of this approach to reliably detect a relatively small object (such as a hand) pressed against the leading edge of a closing door provides a poor comparison to the performance of an IR light curtain.

The other highly discussed technology is using a camera (or cameras) to monitor the scene. However, if only one camera is used, then like a person with only one eye, the system has no perception of depth.  Distinguishing between a shadow or a black shoe crossing the threshold becomes a matter of deep AI and high-level processing.  With enough computing power it may be possible to give reliable results, however in reality, as autonomous vehicle manufacturers and others have found, a single camera sensor is not a realistic solution, and certainly not for a safety function.  What a single camera is good at sensing is changing patterns.  In this way it can provide good tracking of objects and in doing so, with the appropriate processing, providing a relatively effective 3D detection where a 99% accuracy is acceptable, unlike the primary safety function of 2D, where 100% accuracy even for stationary objects is required.  Of course, such systems will still struggle with shadows and to maintain their function during periods of low lighting conditions (e.g. lighting failure) unless they have their own light source.  For a camera solution to work reliably it would need its own light source, stereo cameras, a wide viewing angle and significant processing.  This is technically possible today but creating a product that does all this at a market acceptable price has yet to occur and indeed may never do so. Indeed, our laboratory testing has shown how easily defeated camera or ToF systems are when exposed to the ASME 17.1:2019 test target.

What is the solution to ASME17.1:2019?

Currently the only solution on the market that meets the ASME 17.1:2019 code at an acceptable price point is a system that uses the reliable technology of IR light curtains for 2D detection combined with well proven radar technology for 3D.  By using radar for 3D, which detects reflection caused by matter rather than the light, this solution sidesteps all the complexity and unreliability of other solutions.  Whilst the technology may not be as exciting, its simplicity ensures 100% accurate detection in all conditions.