The way we walk says a lot about us, perhaps even more than we would like. With the help of artificial intelligence (AI), movement patterns can be detected in the way each individual walks that can help prevent some diseases. But also to identify people, although the level of reliability of these tools is not yet definitive. Scientists from the University of Adelaide (Australia) and the Johannes Gutenberg University (Germany) have developed a new method that improves registrations, reaching an accuracy rate of 89% in some tests.
The approach of the researchers, who have published their conclusions in the journal Interfacepublished by the Royal Society, consists of analyzing the forces that are applied to the ground when walking. To do this, they experimented with 700 volunteers from various countries who walked, with shoes and barefoot, on a ten-meter-long platform capable of measuring the pressure exerted on it and the two-dimensional coordinates of the center of pressure. The models trained with people in their own shoes, those responsible for the research say, achieved an accuracy of over 89%. However, in other tests the rate was 52%, showing that there is still a long way to go for these tools to be reliable.
In any case, it is a technology that has gained popularity in the last five years, but which lacks development. “There is still no scientific consensus about whether each person has a completely genuine way of moving, which allows an individual to be identified without any doubt,” says Lorena Jaume-Palasí, advisor to the European Parliament and the Max Planck Institute on AI. “That is something that is mentioned in the study, but it is not said whether or not its contribution contributes to this key scientific debate,” warns this expert, who has founded The Ethical Tech Society, a multidisciplinary study group on the consequences social aspects of automatic systems.
The characteristic features of the way of walking, ranging from the type of step or the speed of movement to small tics or gestural manias, including the inclination or momentum of the walk, can be recorded with a series of sensors such as cameras. infrared or motion, electromyography cameras or force platforms. With that data and with the help of machine learning techniques, models can be built to analyze patterns. “Gaits are a personal story written by the body, a tool for understanding biological identity in medicine and security,” Kayne A. Duncanson, from the University of Adelaide, and her colleagues say in their work.
The research carried out on this question focuses on two main applications. The first is the medical one. “Human gait is a multifaceted behavior that manifests itself through the complex interaction of multiple organic systems. In the field of medical care, the objective is to use gait as a personal functional marker to help in the management of neurological and musculoskeletal pathologies,” maintain the authors of the study.
The second major application is a bit more disturbing: security. The goal is to identify individuals by their movement patterns. It would be enough to associate one of these genuine patterns (if any) with the identity of a subject so that an AI system can recognize it every time it is shown moving images of the person in question. The researchers explain the complexity of this task: “Gait recognition requires modeling at the individual level to detect characteristics that vary between subjects, but that remain constant in individuals over time. Therefore, most studies focus on developing complex multivariate models, such as deep neural networks, to separate gait identification features from body appearance-related features.”
Imprecise technology
The results of the study should be read more on a theoretical than a practical level. By opting for measurement techniques that involve stepping on a surface with sensors, instead of infrared cameras or 3D visualization, its status as a laboratory experiment is made explicit. On the other hand, an accuracy rate of 89% is, according to Jaume-Palasí, unacceptable in fields such as medicine or security elements. “Would you ride an elevator that works well 89% of the time?” says the expert ironically. “An acceptable level of precision depends on the geographic context or the ethical severity in which you are operating,” he adds.
The authors themselves acknowledge that the results of their laboratory tests vary significantly depending on factors such as “footwear, walking speed, body mass, sex, height, and possibly other time-dependent factors.” Jaume-Palasí includes in this section “the state of mind, the type of clothing worn, whether there are people around who interfere with walking, whether objects are being dragged or rolled, and the moment in which the measurement occurs” (not specified). It’s the same on a Sunday walk as it is in a subway station during rush hour).
Automatic gait recognition tools for safety purposes have been used for two decades, especially in Anglo-Saxon countries. The way we walk has been admitted in trials as forensic evidence in both the United Kingdom and the United States and Canada. The first recorded occurrence was in 1839 in London, when Thomas Jackson, who had a bowed left leg and walked with a limp, was identified by a witness, George Cheney. “I recognize him by the way he walks,” he declared. There are more recent cases, supported by AI tools, although there are also those who dismiss this evidence due to lack of scientific rigor.
