A side bending detection method and system based on a multi-semantic segmentation network

Abstract

The present application relates to a kind of based on multi semantic segmentation network's side bending detection method and system, wherein side bending detection method includes respectively inputting a to be identified image into the first segmentation subnetwork and the second segmentation subnetwork in a trained multi-view semantic segmentation network to carry out semantic segmentation and obtain the first subnetwork segmentation graph and the second subnetwork segmentation graph, to two subnetwork segmentation graph line uncertainty minimization competition and synthesize a segmentation prediction image;The segmentation prediction image is carried out human key part identification and key point matching, obtains multiple matched key points;According to multiple matched key points, calculate index parameter, and according to index parameter, output prediction result, through multi-subnet respectively focus on learning the image features of segmentation graph, then combine uncertainty minimization competition and extract the important features of two images, synthesize a high-precision segmentation prediction image, so that it can be identified image can be high-precision segmentation, realize high-precision side bending detection.

Images