2-D Omnidirectional Human Body Attribute Fusion Recognition System

Abstract

Traditional face or body attribute recognition can solely train and understand the specific attributes marked according to the marked content of the dataset. Most recognition systems on the market integrate specific attribute recognition algorithms. The lack of corresponding attribute data fusion leads to the problem of high diversity but poor recognition accuracy. This paper designs a recognition system based on data fusion and attribute fusion of two-dimensional images, which can fuse subtle data during training, and fuse the recognition results of overlapping attributes, to achieve multi-directional recognition of human subtle attributes, make up for the shortcomings of single recognition algorithm, and obtain more comprehensive and accurate recognition results. The experimental results show that the recognition system. Firstly, the overall recognition accuracy is above 83.5%. Secondly, it can make up for the problems of high accuracy but poor diversity of single recognition algorithm and poor accuracy of multi-feature recognition algorithm.