InvKA: Gait Recognition via Invertible Koopman Autoencoder

TL;DR

InvKA introduces an interpretable gait recognition method using Koopman operator theory and a reversible autoencoder, significantly reducing computational cost while maintaining high accuracy.

Contribution

The paper proposes a novel gait recognition approach leveraging Koopman operator theory and invertible autoencoders for improved interpretability and efficiency.

Findings

Reduces computational cost to 1% of state-of-the-art methods

Achieves 98% recognition accuracy on non-occlusion datasets

Provides physically meaningful gait features via Koopman operator

Abstract

Most current gait recognition methods suffer from poor interpretability and high computational cost. To improve interpretability, we investigate gait features in the embedding space based on Koopman operator theory. The transition matrix in this space captures complex kinematic features of gait cycles, namely the Koopman operator. The diagonal elements of the operator matrix can represent the overall motion trend, providing a physically meaningful descriptor. To reduce the computational cost of our algorithm, we use a reversible autoencoder to reduce the model size and eliminate convolutional layers to compress its depth, resulting in fewer floating-point operations. Experimental results on multiple datasets show that our method reduces computational cost to 1% compared to state-of-the-art methods while achieving competitive recognition accuracy 98% on non-occlusion datasets.