<p>Parkinson’s disease (PD) is the second most common neurodegenerative disorder, primarily affecting motor functions. Abnormal gait patterns encompass a wide range of motor symptoms commonly used for diagnosis from observational analysis. Recently, computer-based approaches have emerged as non-invasive alternatives to support the quantification of these abnormal patterns, particularly from video analysis. Given the high dimensionality and limitation of data, compact temporal covariance descriptors have been employed to capture discriminative spatio-temporal dynamics in gait recordings. From this, Riemannian geometry and support vector machines (SVM) with an RBF kernel have demonstrated strong performance to discriminate PD-related descriptors in prior works. In this work, we investigate whether SVM performance over these covariance descriptors can be further improved by using a quantum kernel. We transformed gait videos of healthy individuals and patients with PD into covariance matrices and compared the SVM performance with and without a quantum kernel. Four different kinds of quantum kernels were benchmarked as part of this study. The results did not conclusively demonstrate the superiority of the quantum kernel over the RBF kernel. However, using the common spatial pattern for feature extraction improved the classification performance compared to the state of the art.</p>