Vision-based biometric authentication system using human gait analysis (Vision GaiA)

• Main Authors: Infante, Richo Paulo M., Salcedo, Gianne Lauren O., Tan, Ralph Gerard M.
• Format: text
• Language: English
• Published: Animo Repository 2009
• Subjects: Walking; Pattern recognition systems; Identification > Automation; Biometric identification
• Online Access: https://animorepository.dlsu.edu.ph/etd_bachelors/8667
• Institution: De La Salle University

Gait refers to the particular way or manner of moving on foot. The gait of a person refers to the unique manner of his or her movement while walking. Recently, gait analysis has been examined for the purposes of biometrics and rehabilitation and is being researched further for identification and authentication purposes. By studying gait analysis for authentication and identification, the percentage of raising the possibility of enormous security risks and threats is lessened. Using the video frame sequences obtained from a video camera, the Vision-Based Biometric Authentication System using Human Gait Analysis (Vision GaiA) extracts features that can be obtained from the walking person and uses them to authenticate the person by checking if his or her gait template exists in the database. To extract these features, the location of the walking person from the video frame sequence is first obtained by segmenting the foreground from the background and is binarized to obtain the binary silhouette. The gait features extracted are the aspect ratio of the silhouettes' bounding box, the height and width signal of the bounding box, the computed silhoutte centroid coordinates, and as well as the computed average frequency of the person's gait cycle. Authentication is done by applying Euclidean distance matching between the query gait data and the reference gait template of the person. Using nearest-neighbor rule, only the instance with the highest similarity value among the population of the reference gait template is considered. The highest similarity value is then compared with the optimal system threshold of 85.5%. This is set by the Equal Error Rate Test after computations of error rates generally yielding favorable authentication result of 8%.