A gait analysis-based personal identity recognition method is implemented by a computer system, and includes: acquiring multiple pieces of training feature data obtained by sensing vibration of an inertial sensing board while a specific person walks on the inertial sensing board for multiple steps; training a Siamese neural network based on the pieces of training feature data in an input manner that two piece of training feature data are inputted to the Siamese neural network each time and adjusting its shared weights to establish a gait similarity estimation model; obtaining feature data based on inertial sensing data that is generated by sensing the inertial sensing board while a subject walks on the inertial sensing board for one step; and identifying whether the subject is the specific person based on a predetermined threshold, and on a plurality of similarity values each associated with the feature data and a corresponding one of the piece of training feature data and estimated by the gait similarity estimation model. |