Gait recognition is a biometric method that uses sensor data to recognize people based on body shape and walking styles. Gait data is acquired from video images, inertial sensors, or sensors in the environment. The possible uses are diagnostic, but most of the research is aimed at recognizing people for forensic use.
The authors survey 198 papers, spanning from 1994 to 2018. In the past 20 years, the introduction of new sensors and the creation of large datasets have sped up this research topic, both in model-based methods (which extract features from a body model) and model-free methods. The survey covers the steps of data acquisition, feature representation, and dimension reduction, using videos, accelerometers, floor sensors, and radars. Moreover, the authors report on classifiers using statistics and artificial intelligence (AI) methods. The presentation concludes with an agenda of seven open points to address.
Researchers will find many ideas for improving both gait recognition and incomplete theories. Readers working in security will find the survey’s analysis of the vulnerabilities of gait-based biometrics interesting; factors such as body type, clothes worn, and variation in walking speed affect results. Readers looking for what is available today will find indications about the datasets and about recognition accuracy, “with misclassification rates less than 0.15 [percent].”