Part definition and selection for part-based speed invariant gait recognition

Walking speed change is one of the most common cofactor that affects the gait signature. The intra-class variations increases by the changes of walking speed and the training data set fails to model the variations. As a consequence, the performance of the gait recognition systems is degraded. Naturally, human reconstruct the model by considering the unaffected body parts to recognize a person when some parts are affect due to the walking speed transition. In this paper, we propose a general feature set based method to recover from the speed transition by using the less affected body parts. The OUISIR Treadmill Database A is used to show the effectiveness of the proposed method.