An identity recognition method for determining whether a user belongs to a group consisting of P number of members based on gait recognition, where P is a positive integer, the identity recognition method to be implemented by an identity recognition system and comprising:
- storing P number of training data sets that respectively correspond to the P number of members and that are respectively related to gaits of the P number of members, each of the P number of training data sets including Q pieces of training feature data that respectively correspond to Q number of steps taken by a corresponding one of the P number of members on at least one sensor mat and that are obtained based on an inertial sensing result generated by the at least one sensor mat which is capable of detecting vibrations with respect to multiple inertial features, where Q is an integer greater than one;
- obtaining a gait similarity estimation model that is established by training a Siamese neural network using the P number of training data sets in such a manner that, for each round of training, two pieces of training feature data, which are selected from among the P number of training data sets as two pieces of input data and which respectively correspond to two different steps, are simultaneously inputted into the Siamese neural network to adjust weights used by the Siamese neural network;
- obtaining at least one piece of inertial data that corresponds to a step taken by the user on the at least one sensor mat and that is generated by the at least one sensor mat which detects vibrations resulting from the step with respect to the inertial features;
- generating, based on the at least one piece of inertial data, a piece of feature data that is related to gait of the step taken by the user;
- based on the piece of feature data and the P number of training data sets, generating P number of evaluation results that respectively correspond to the P number of training data sets by using the gait similarity estimation model, wherein, with respect to each of the P number of training data sets, the corresponding evaluation result includes Q number of similarity values each between the piece of feature data and a respective one of the Q pieces of training feature data of the training data set;
- determining whether the user belongs to the group consisting of the P number of members based on the P number of evaluation results and a predefined threshold;
- and
- unlocking an access control system by outputting a signal of identity recognition which indicates a positive result of identity recognition to the access control system when it is determined that the user belongs to the group,
- wherein each of the at least one sensor mat includes multiple inertial sensors that are arranged in an array on a surface of the at least one sensor mat, each of the inertial sensors detecting accelerations and angular velocities in three axes that represent the inertial features to generate six inertial sensing signals, and
- wherein, in the storing P number of training data sets, each of the Q pieces of training feature data of each of the P number of training data sets includes six 4×R feature matrices that respectively correspond to the inertial features, and that were obtained by performing sampling at a sampling rate of R on each of four groups of the six inertial sensing signals which were generated by four inertial sensors that are closest, among all the inertial sensors, to a spot at which the corresponding step taken by the corresponding member fell on the at least one sensor mat.