Human gait recognition by shoulder movement's Doppler signature using SVM classifier

• Main Authors: Elyass, Sinan M., Abed, Aqeela N., Hussei, Jabar S., Mahmoud, Aseel G., Ziboon, Hadi T., Saeed, Thamir R.
• Format: Conference Proceeding
• Language: English
• Subjects: Biometric recognition systems; Classifiers; Doppler radar; Euclidean geometry; Gait; Gait recognition; Human motion; Line of sight; Linearity; Privacy; Radar signatures; Shoulder; Support vector machines; Voice recognition; Weather
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0027556

Doppler radar has been used as a sensing device for human gait identification because it possesses the ability to work in adverse weather and penetrate invisible barriers with maintaining privacy. Doppler explains the velocity of the target's motion, and the micro-Doppler explains the micro-motions of its moving parts. The Doppler and micro-Doppler signatures are used to classify the target. Many studies and approaches were presented for human recognition like fingerprints, retina, palm, face, and voice. The gap of this subject is concentrated on three points, the first one needs the subject's permission and physical attention, the second one is maintaining the privacy. Also, the location of the detector device and the detected human part. This paper introduces a novel approach for human gait recognition based on his shoulder's movement signature and periodic motion to satisfy the recognition and to fill this gap. The micro-Doppler of shoulders (or head) is detected by radar, which was fixed in the ceiling. The radar location was decided because the shoulder movement is in-phase with the line of sight (LOS) of the radar signal. The support vector machine (SVM) was used as a classifier. Five people were tested as classes with many patterns for each class related to a different speed and motion status. The Euclidean and Bhattacharyya distances were used to measure the similarity. The Euclidean distance is overperformed the Bhattacharyya distance in this application by 14% caused by its linearity. Simplicity and accuracy can be considered as the distinguishing features of the proposed system in which the gained classification rate is 96.2%.