Real-time continuous gesture recognition system based on PSO-PNN

Abstract

Abstract In the field of natural human–machine interaction, wearable gesture interaction technologies have received considerable attention, particularly continuous gesture (CG) recognition. However, CG faces several challenges, including the impact of motion characteristics on gesture recognition and performance that is not sufficiently robust. Traditional algorithms are highly dependent on samples, thus meeting the requirements of low sample volume and high accuracy simultaneously is challenging. To address these challenges, we propose a real-time CG recognition system based on particle swarm optimization (PSO) and probabilistic neural network (PNN). This system employs principal component analysis for signal dimensionality reduction to alleviate computational burden and uses K-means clustering and Pearson correlation coefficient to extract optimal features for gesture classification. In offline gesture recognition experiments involving six CGs, the algorithm achieved an accuracy rate of 97% with a training set of 300 samples and a runtime of just 31.25 ms. Compared to other five algorithms, the proposed algorithm improved accuracy by at least 9% and reduced the runtime by 40.475 ms. Moreover, gesture recognition experiments were conducted using different datasets, with the PSO-PNN algorithm achieving an average recognition rate of 90.17%, at least 9.84% higher than other algorithms. Finally, in experiments on online CG control for robots in complex environments, the PSO-PNN demonstrated real-time performance of 28.56 ms and a task completion rate of 90.67%, validating the feasibility of PSO-PNN. This research provides a substantial theoretical and technical foundation for the ongoing enhancement and application of CG recognition technology.