Multi-gait Synthesis Based on Convolutional Neural Networks

Multi-gait synthesis aims to describe the mapping relationship between single gaits and multi-gaits, allowing predictions of multi-gait forms based on the single gait forms of each participant. Our goal is to utilize a dual-branch input pipeline, where each separate branch learns the gait features of each individual, aggregating the gait sequences of two different individuals to generate a complete dual-person gait sequence. Experiments conducted on our lab’s self-collected multi-person gait dataset have shown that our model can generate satisfactory multi-person gait sequences. Furthermore, testing the generated gait images with the GaitSet model for gait recognition demonstrates that the image quality produced by our model is acceptable.