How accurately do face recognition software tools identify people of varied sex, age and racial background? According to a new study by the National Institute of Standards and Technology (NIST), the answer depends on the algorithm at the heart of the system, the application that uses it and the data it’s fed – but the majority of face recognition algorithms exhibit demographic differentials. A differential means that an algorithm’s ability to match two images of the same person varies from one demographic group to another. Results captured in the report, Face Recognition Vendor Test (FRVT) Part 3: Demographic Effects (NISTIR 8280), are intended to inform policymakers and to help software developers better understand the performance of their algorithms. Face recognition technology has inspired public debate in part because of the need to understand the effect of demographics on face recognition algorithms. “While it is usually incorrect to make statements across algorithms, we found empirical evidence for the existence of demographic differentials in the majority of the face recognition algorithms we studied,” said Patrick Grother, a NIST Computer Scientist and the Report’s Primary Author, “While we do not explore what might cause these differentials, this data will be valuable to policymakers, developers and end users in thinking about the limitations and appropriate use of these algorithms.” The study was conducted through NIST’s Face Recognition Vendor Test (FRVT) program, which evaluates face recognition algorithms submitted by industry and academic developers on their ability to perform different tasks. While NIST does not test the finalized commercial products that make use of these algorithms, the program has revealed rapid developments in the burgeoning field. The NIST study evaluated 189 software algorithms from 99 developers – a majority of the industry. It focuses on how well each individual algorithm performs one of two different tasks that are among face recognition’s most common applications. The first task, confirming a photo matches a different photo of the same person in a database, is known as ‘one-to-one’ matching and is commonly used for verification work, such as unlocking a smartphone or checking a passport. The second, determining whether the person in the photo has any match in a database, is known as ‘one-to-many’ matching and can be used for identification of a person of interest. To evaluate each algorithm’s performance on its task, the team measured the two classes of error the software can make – false positives and false negatives. A false positive means that the software wrongly considered photos of two different individuals to show the same person, while a false negative means the software failed to match two photos that, in fact, do show the same person. Making these distinctions is important because the class of error and the search type can carry vastly different consequences depending on the real-world application. “In a one-to-one search, a false negative might be merely an inconvenience – you can’t get into your phone, but the issue can usually be remediated by a second attempt,” Grother said, “But a false positive in a one-tomany search puts an incorrect match on a list of candidates that warrant further scrutiny.” What sets the publication apart from most other face recognition research is its concern with each algorithm’s performance when considering demographic factors. For one-to-one matching, only a few previous studies explore demographic effects; for one-to-many matching, none have. To evaluate the algorithms, the NIST team used four collections of photographs containing 18.27 million images of 8.49 million people. All came from operational databases provided by the State Department, the Department of Homeland Security and the FBI. The team did not use any images ‘scraped’ directly from internet sources such as social media or from video surveillance. The photos in the databases included metadata information indicating the subject’s age, sex, and either race or country of birth. Not only did the team measure each algorithm’s false positives and false negatives for both search types, but it also determined how much these error rates varied among the tags. In other words, how comparatively well did the algorithm perform on images of people from different groups? Tests showed a wide range in accuracy across developers, with the most accurate algorithms producing many fewer errors. While the study’s focus was on individual algorithms, Grother pointed out five broader findings: For one-to-one matching, the team saw higher rates of false positives for Asian and African American faces relative to images of Caucasians. The differentials often ranged from a factor of 10 to 100 times, depending on the individual algorithm. False positives might present a security concern to the system owner, as they may allow access to impostors. Among U.S.-developed algorithms, there were similar high rates of false positives in one-to-one matching for Asians, African Americans and native groups (which include Native American, American Indian, Alaskan Indian and Pacific Islanders). The American Indian demographic had the highest rates of false positives. However, a notable exception was for some algorithms developed in Asian countries. There was no such dramatic difference in false positives in oneto-one matching between Asian and Caucasian faces for algorithms developed in Asia. While Grother reiterated that the NIST study does not explore the relationship between cause and effect, one possible connection, and area for research, is the relationship between an algorithm’s performance and the data used to train it. “These results are an encouraging sign that more diverse training data may produce more equitable outcomes, should it be possible for developers to use such data,” he said. For one-to-many matching, the team saw higher rates of false positives for African American females. Differentials in false positives in one-to-many matching are particularly important because the consequences could include false accusations. (In this case, the test did not use the entire set of photos, but only one FBI database containing 1.6 million domestic mugshots.) However, not all algorithms give this high rate of false positives across demographics in one-to-many matching, and those that are the most equitable also rank among the most accurate. This last…

Microsleep, distraction, a seatbelt left undone – many things that happen inside a vehicle can have far-reaching consequences. To avert critical driving situations and possibly also accidents, it is planned that cars will in the future use their sensors not simply to monitor the road but also the driver and other passengers. For this purpose, Bosch has developed a new interior monitoring system featuring cameras and artificial intelligence (AI). “If the car knows what its driver and occupants are doing, driving will become safer and more convenient,” said Harald Kroeger, a member of the Robert Bosch GmbH board of management. The Bosch system may go into production in 2022. In that year, the EU will make safety technology that for example warns drivers of drowsiness and distraction a standard feature in new vehicles. The EU Commission expects that, by 2038, their new safety requirements for vehicles will save more than 25,000 lives and help prevent at least 140,000 severe injuries. By keeping an eye on what is happening inside the car, it is hoped that a fundamental problem of self-driving cars will be solved. If responsibility for driving is to be transferred to the driver again following an automated drive on the freeway, say, the car needs to be sure that the driver is neither sleeping, nor reading the newspaper, nor writing e-mails on their smartphone. A smart camera constantly monitors the driver At 50 kph, a vehicle will cover 42 meters completely unsupervised if the driver dozes off or looks at their smartphone for just three seconds. Many people underestimate the associated risk. International studies state that nearly one in ten accidents are caused by distraction or drowsiness. This has prompted Bosch to develop an interior monitoring system that detects and alerts to this danger and provides driving assistance. A camera integrated in the steering wheel detects when drivers’ eyelids are getting heavy, when they are distracted, and when they turn their head toward their passenger or the rear seats. Thanks to AI, the system draws the right conclusions from this information – it warns inattentive drivers, recommends a break if they are getting tired, or even reduces the speed of the vehicles – depending on the automaker’s wishes, and also on legal requirements. “Cameras and AI will turn the vehicle into a life-saver,” Kroeger continued, To achieve this, Bosch engineers have used intelligent image-processing algorithms and machine learning to teach the system to understand what the person in the driving seat is actually doing. To take the example of driver drowsiness, the system is trained using recordings of real driving situations, and on the basis of recordings of eyelid position and eyeblink rate, learns how tired the driver really is. This allows it to give an alert that is appropriate to the situation, and to use the driver assistance systems to intervene. Warning systems that sound the alert in the case of distraction and drowsiness will be so important in the future that NCAP, the European New Car Assessment Program, will include them in the roadmap for the Euro NCAP assessment for vehicle safety by 2025. On the subject of monitoring, only the software in the vehicle itself evaluates the information provided by the interior monitoring system – the information is neither saved nor passed on to third parties. Like a relay race: Responsibility for steering passes from car to driver and back At the latest when cars start driving automatically, it is obvious how important it is that they understand their drivers. Once driving is automated, cars will drive along free ways without driver intervention. However, they will also have to be able to hand back control to their drivers – in tricky situations such as construction zones, or when the exit ramp is drawing near. Drivers have to be able to safely take the wheel again at any time during the automated driving phase, and the camera makes sure they don’t fall asleep. If their eyes remain closed for a prolonged period, an alarm is sounded. The system also interprets camera recordings to establish what drivers are currently doing, and how ready they are to respond. The transfer of driving responsibility is then timed accordingly. “Bosch driver observation will be essential for safe automated driving,” Kroeger commented. When the car keeps its camera eyes open But the new Bosch system keeps its eye not only on the driver, but also on all the other passengers, whether next to or behind the driver. For this purpose, a camera mounted above or below the rear-view mirror monitors the entire passenger compartment. It notices whether children on the rear seats have carelessly unfastened their seat belts, and warns the driver. If someone sitting in the back is leaning too far forward, at an angle, or with their feet up on the seat next to them, the airbags and belt tensioner will not be able to protect them properly in an accident. The interior monitoring camera can tell what position they are sitting in and set the airbags and belt tensioner to ensure the best possible protection. The interior monitoring system also prevents the passenger seat airbag from being deployed if a baby’s carrycot is on the seat. On the subject of children, it is a sad fact that parked vehicles can be a death trap for them. In the United States in 2018, they claimed the lives of more than 50 children (source: KidsAndCars. org), either because they had been left in the car for a short while or had clambered in unnoticed. The new Bosch system can recognize this danger and warn parents in a flash by sending a message to their smartphones. In an emergency, it also can alert the emergency services. As the Hot Cars Act currently being debated in the United States shows, legislators are interested in technology solutions to address this challenge. A camera for more convenience The new Bosch system also means more driving convenience. The interior monitoring camera can tell who…