Research on optimization of face recognition algorithm in smart campus driven by multimodal feature fusion: construction of hybrid model based on dynamic weight allocation

In order to solve the problems of low recognition accuracy and poor robustness of face recognition technology in smart campus in complex environment, a hybrid model optimization method based on multi-modal feature fusion (MDFF) and dynamic weight assignment (DWA) is proposed. In this model, texture, shape and dynamic features of human face are fused, and texture, shape and dynamic features are extracted by local binary pattern (LBP), Canny edge detection and optical flow method respectively, and the weights of each feature are dynamically adjusted according to the distinguishing ability of features and scene conditions by DWA strategy. The experiments are carried out with CASIA-WebFace, LFW and self-built campus data sets. The results show that the recognition accuracy of the proposed method in Top-1 reaches 92.0%, 96.5% and 91.4% respectively, which is significantly better than the traditional method and the single-mode method. DWA strategy can adaptively adjust the weight distribution in different scenes, especially in backlight, occlusion and dynamic fuzzy scenes, which improves the robustness of the model.