We propose a novel bolt-on module capable of\nboosting the robustness of various single compact 2D gait\nrepresentations. Gait recognition is negatively influenced by\ncovariate factors including clothing and time which alter\nthe natural gait appearance and motion. Contrary to traditional\ngait recognition, our bolt-on module remedies this by\na dedicated covariate factor detection and removal procedure\nwhich we quantitatively and qualitatively evaluate. The\nfundamental concept of the bolt-on module is founded on\nexploiting the pixel-wise composition of covariate factors.\nResults demonstrate how our bolt-on module is a powerful\ncomponent leading to significant improvements across gait\nrepresentations and datasets yielding state-of-the-art results
Loading....