A method of tracking an object using a CNN including a tracking network, and a device using it {METHOD FOR TRACKING OBJECT BY USING CONVOLUTIONAL NEURAL NETWORK INCLUDING TRACKING NETWORK AND COMPUTING

摘要

PROBLEM TO BE SOLVED: To provide a method for tracking an object by using a CNN.;SOLUTION: A method comprises steps of: on a device, acquiring a feature map from a current video frame of a video, and instructing an RPN to generate information on a proposal box (PB) for an object; generating, from a previous state vector of a previous bounding box (BB) for the object located on a previous video frame, an estimated state vector of the previous BB by using a Kalman filter algorithm (KFA); generating an estimated BB in the current video frame tracked from the position of the previous BB for the estimated state vector; determining a specific PB with reference to the estimated BB; instructing an FCN to apply operations to the feature map to output a position-based score map; instructing a pooling layer to calculate a class score and a regression delta for a specific PB; generating a current BB for an object on the current video frame; and adjusting the current BB by using the KFA.;SELECTED DRAWING: Figure 2;COPYRIGHT: (C)2020,JPO&INPIT