Mercedes’ New Idea For Detecting Occupants Should Spare Third Owners From At Least One Annoying Dash Light

Having owned a BMW from the mid-2000s, I’m acutely aware of the existence of seat occupancy sensors. These sensors are embedded into the seats of pretty much every car on sale today, and detect whether someone’s, well, occupying the seat. These sensors then relay that information back to the car, informing the computer about stuff like which seat belts should be buckled, and in some vehicles, whether to arm the passenger-side airbag.

A new patent filed by Mercedes-Benz is reinventing the idea of seat occupancy sensors by throwing them out entirely in favor of a new method for detecting how many people are inside a vehicle, and who’s sitting where.

Before diving into Mercedes’ idea, it’s worth explaining how seat occupancy sensors normally work. FSRTEK, a producer of these sensors, explains how exactly they detect whether someone’s ass is in a seat:

The car seat sensor is a flexible, contact sensor. Because the contact is composed of silver paste, ink and other components, it is evenly dispersed on the surface of the force. When external pressure is generated, it will trigger a signal to achieve detection. There are a variety of seat-related purposes such as whether there are people in the seat, whether the sitting posture is safe, whether the seat belt is standardized, and so on.

Some modern seat occupancy sensors are fancier than others, able to detect the weight of whoever is sitting in the seat to determine how the car’s airbag will function. These fancier sensors are called occupant classification systems (OCS). Here’s how they differ, according to the collision repair experts at I-CAR:

OCS have been an integral part of supplemental restraints systems (SRS) since the early 2000s. They are an important part of making sure the air bag properly protects the passenger. Designed for detecting passenger presence and weight, they may also be able to detect if a child or small adult is in the seat. This is vital to making sure the passenger air bag deploys only under the correct circumstances.

The system typically consists of a weight sensor, seat position sensor, and a control module. The weight sensor may be a bladder-style sensor, strain gauges, or a pressure-sensitive mat. Often, there is a switch for disabling the passenger air bag.

OCS devices are the reason you don’t see those keyholes for turning the passenger airbag on and off on newer cars—the car does the math for you, so if you get in a crash with your kid or some other lightweight individual in the passenger seat won’t get injured by the airbag’s detonation.

It’s a pretty good idea, and one that most people will never even know exists. But if you’re someone like me, you know seat occupancy sensors can be a huge pain to fix when they go bad. Back in 2020, when I still owned my 2008 BMW M5, the passenger-side occupancy sensor went bad, triggering the car to display an airbag light and a seatbelt alert.

The only way to fix the sensor is to replace it, which would’ve required the entire seat to come apart—something I couldn’t do myself, and would’ve cost me over $1,000 to pay someone to do for me. The only other solution was to buy a little dongle from eBay that plugs into the car under the seat, which would emulate a working sensor, essentially tricking the car into thinking there was always someone sitting and buckled in the passenger seat. The dongle cost me about $57, so that’s the route I went. It only takes a few minutes of searching the web to see dozens of others dealing with this same problem on all sorts of cars.

This Mercedes Patent Chucks That Whole System In The Trash

The idea, published to the World Intellectual Property Organization last week, describes a method for detecting seat occupancy by using a camera rather than any physical sensors. It works by taking a picture of the cabin from the rear-view mirror area, then running the photo through an algorithm to determine whether people are sitting in seats. Here’s how it’s described by the WIPO page:

The invention relates to a method and/or a device for detecting a seat occupancy in a vehicle, characterized by the following method steps (a to d): a) capturing at least one image with a first resolution of pixels by an image sensor, wherein the image sensor is divided into a grid of polygons, b) summarizing those pixels of the captured image that lie within a polygon to form a sum pixel value and summarizing all sum pixel values of all polygons to form an aggregated image, c) examining the aggregated image by at least one of two predetermined image recognition algorithms (c1), c2)), c1) with a first image recognition algorithm (c1)) to detect a presence of a person and optionally to terminate the first image recognition algorithm (c1)) if a person has been detected, and c2) with a second image recognition algorithm (c2)) for detecting a vehicle seat and optionally for terminating the second image recognition algorithm if a vehicle seat without a person using it has been detected, and d) optionally recording at least one further image with a second resolution of pixels by the image sensor (15) if neither a person nor a vehicle seat has been detected, and optionally repeating the method steps (a/b) to c1/c2/d)) until one of the two image recognition algorithms (c1), c2)) is or is terminated.

While seat occupancy sensors aren’t particularly heavy, having, say, five of them installed into seats throughout a car is still heavier than having one small camera facing the cabin. And that camera is far easier and cheaper to replace. Mercedes doesn’t mention either weight or ease of replacement, but rather the accuracy of such a system over traditional sensors as the main benefit, even when using low-quality photos. From the patent:

The advantage of the invention is, in particular, that due to the aggregation of pixels of a polygon or multiple polygons of the recorded image into a single pixel value or sum pixel value of the polygon and the aggregation of all sum pixel values ​​of all polygons into an aggregated image and their processing using two different image recognition algorithms, even low-quality images are sufficient to detect whether a person is sitting in the vehicle or not. The lower quality of the images can be sufficient for the respective specific image recognition algorithm to detect whether or not a person is in the vehicle or on a seat. In particular, the two differently configured image algorithms allow lower-quality images to be evaluated more quickly and reliably with regard to seat occupancy in the vehicle. If the quality of the images is not sufficient for either of the two image algorithms, repeating process steps a)/b) to c1)/c2/d) by taking images with a higher quality, with a better resolution, in particular a higher pixel resolution and/or a higher raster resolution, and processing them using one or both of the image recognition algorithms can make it possible to detect more quickly and reliably whether a person is sitting on a vehicle seat or not.

Mercedes even mentions a feature that could call out an already-identified occupant for sitting in their seat incorrectly—something that a normal occupancy sensor might not be smart enough to do:

In addition, a higher resolution can be used by the image recognition algorithms to carry out an additional analysis, for example to more closely identify a person who has already been detected and/or to determine whether the seating position of the detected person is sufficient to meet safety functions and/or safety requirements in the vehicle.

While there are some obvious privacy concerns with having a camera pointed towards the occupants of cars, it’s worth highlighting that Mercedes has been putting cabin-facing cameras in its cars for years now. So seeing something like this isn’t exactly surprising.

As a chronic third-owner of depreciated, crumbling German luxury cars, I’m all for this idea. If letting a Mercedes algorithm analyze my face means I don’t have to deal with sensors going bad, then so be it.

Top graphic images: Mercedes-Benz; Logitech