Target detection method, device, equipment and storage medium

Abstract

The application discloses a target detection method, device, equipment and storage medium, and relates to computer vision technology of artificial intelligence. The method includes: extracting image features of an input image; predicting the image features to obtain a size image, an angle image, and a center point image; performing rotational convolution on the image features, and outputting a first classification result, the rotational convolution comprising Using a rotation matrix to perform rotation sampling on the image feature, and using a convolution kernel to convolve the rotation sampling matrix obtained by the rotation sampling; performing center point focus processing on the image feature area in the image feature, and outputting the second classification As a result, the central point focusing process is used to weight the central point of the detection target based on a Gaussian distribution. This method can improve the accuracy of object detection.

Description

technical field [0001] The present application relates to artificial intelligence computer vision technology, and in particular to a target detection method, device, equipment and storage medium. Background technique [0002] With the development of deep learning, object detection technology has made significant progress. Object detection methods based on deep neural networks have quickly replaced traditional methods and become an indispensable key technology in the fields of smart security, home furnishing, and travel. [0003] In the related art, there is provided a target detection method that uses a convolution kernel to extract image features. For example, the image to be detected is an image of 5 pixels*5 pixels, and the convolution kernel is an image of 3 pixels*3 pixels. Using the convolution kernel and Convolute the 3-pixel*3-pixel block in the upper left corner of the image to be detected, and then slide the convolution kernel to the right once to convolve with eac...

AI Technical Summary

Problems solved by technology

[0004] In the scene where detection targets are rotated at any angle and densely arranged, for example, when detecting densely arranged commodities on supermarket shelves, the two detection targets are rotated by 45° and are next to each other. Using the method in the related art, sliding In the process of convolution kernel, if no block can cover a single detection target, the convolution kernel will not be able to extract the accurate features of the detection target, and the result of target detection will be inaccurate.

Images