Reservoir Computing in Rehabilitation Video Analyses
Dongping Yang, Tie Jun Xu, Feng Lin · 2023
Spatio-temporal information processing is fundamental in rehabilitation video analyses. Current strategies for spatio-temporal pattern recognition usually involve explicit feature extraction followed by feature aggregation, which requires a large amount of labeled data. There are two major challenges for recognizing complex spatio-temporal patterns. One is on extracting the representative features; and the other is on extracting the temporal structure, in particular, the temporal order, of image sequences. Reservoir Computing (RC) offers a promising tool by enabling automatic and accurate analyses of complex movement patterns in rehabilitation videos. Unlike traditional machine learning techniques, RC does not require explicit training of the reservoir connections, which is less data- demanding, more resistant to overfilling and easier to implement. The unique properties make it an attractive option for analyzing rehabilitation video data, where patients may experience changes in their movement patterns over time with a large data flow for processing. By analyzing new video data, the reservoir can adapt and refine its output to reflect the patient’s current conditions.