Object detection method and device based on reconfigurable network

Abstract

A method for learning parameters of an object detector based on a CNN adaptable to customer's requirements such as KPI by using a target object estimating network and a target object merging network is provided. The CNN can be redesigned when scales of objects change as a focal length or a resolution changes depending on the KPI. The method includes steps of: a learning device instructing convolutional layers to generate a k-th feature map by applying convolution operations to a k-th manipulated image which corresponds to the (k-1)-th target region on an image; and instructing the target object merging network to merge a first to an n-th object detection information, outputted from an FC layer, and backpropagating losses generated by referring to merged object detection information and itscorresponding GT. The method can be useful for multi-camera, SVM(surround view monitor), and the like, as accuracy of 2D bounding boxes improves.

Description

technical field [0001] The present invention relates to a method for learning the parameters of an object detector based on a reconfigurable convolutional neural network (CNN or ConvNet) using an object prediction network and an object integration network to optimize user requirements such as key performance indicators (KPIs) ), more specifically relate to a learning method and a learning device, and use its testing method and testing device, which utilizes a target object prediction network and a target object integration network to learn the parameters of an object detector based on CNN, which includes: ( a) When at least one training image is input, (i) make one or more convolution layers apply one or more convolution operations to at least one first processed image corresponding to the training image, and output at least one first feature Figure, (ii) make the region candidate network (Region Proposal Network, RPN) use the first feature map to output one or more correspond...

AI Technical Summary

Problems solved by technology

[0004] However, by the convolutional layer of the CNN-based object detector, the input image size is reduced by using the feature map, so it is easy to detect large-sized objects located in the input image, but it is difficult to detect Objects located in the input image with small size

[0005] As another example, a resized image obtained by enlarging the input image can be used to detect small-sized objects, but in this case, the computational load of the object detector increases, thereby reducing the size of the object. Detector performance

Images