JOINT POSITION ESTIMATION DEVICE, JOINT POSITION ESTIMATION METHOD, AND JOINT POSITION ESTIMATION PROGRAM

摘要

In order to increase the accuracy of estimating the position of a joint, this joint position estimation device has a memory and a processor connected to the memory, wherein the processor: performs first deep neural network (hereinafter referred to as DNN) learning using first training data that include input data representing an input image, and that also include correct answer data representing site regions of an animal in the input image; performs second DNN learning using second training data that include input data representing an input image including an image of a site region and an image of a multiple site region that is obtained by combining a plurality of site regions, said second training data also including correct answer data representing joint positions in the site region and in the multiple site region; estimates site regions of an animal that are represented by input image data to be processed, using a first DNN for which a first parameter obtained through the first DNN learning has been set; and, by using a second DNN for which a second parameter obtained through the second DNN learning has been set, estimates a first joint position in a site region estimated using the first DNN, and second joint positions in a multiple site region obtained by combining a plurality of site regions estimated using the first DNN.