The human gait features are regarded as key indicators for disease prevention, fall warning, rehabilitation training, and gait correction applications. How to achieve convenient, efficient, and accurate gait recognition has become a research hotspot. In this work, a novel human gait recognition method is proposed, which effectively combines the merits of the grid-less planar flexible pressure sensors and the multilayer heterogeneous machine learning algorithms. The proposed system has the advantages of multifunctionality, high accuracy, low cost, easy portability, easy scalability, real-time monitoring, and self-learning improvement. In general, pressure sensors can give the information of pressure and location only, yet gait feature extraction involves a complex combination of multizone and time-dependent dynamic process. Therefore, the concept of “plantar patterns” is introduced to bridge the gap between pressure sensing and gait recognition. With the help of neural network algorithm, 12 typical plantar patterns were innovatively extracted from the foot pressure information, and any arbitrary gait could be divided into a combination of one or several typical patterns. Applications including foot shape classification, motion status judgment, and human gait detection were testified experimentally, with an overall prediction accuracy of 97%. This gait recognition method can accurately characterize the dynamic gait features of the human body, providing visualized, accurate, and quantifiable technical support for the early intervention of movement disorders and rehabilitation treatments.