Target detection method and system, computer device and storage medium

Abstract

The invention provides a target detection method and system, a computer device and a storage medium, and the method comprises the steps: obtaining all local images obtained through the segmentation ofan original image, and at least one information enhancement image extracted from the original image; performing feature extraction on each local image and each information enhancement image in parallel in the feature extraction module of each branch to obtain at least one feature extraction image of each branch; splicing and fusing the at least one feature extraction image of each branch according to the corresponding position in the original image to form at least one output feature image; forming a to-be-detected feature image according to the at least one output feature image; and obtaining a target detection result according to the to-be-detected feature image. According to the implementation scheme, the original image is locally divided, feature extraction is performed through the feature extraction module in parallel, so that the detection efficiency is improved, and the target detection effect is improved through multi-dimensional image feature extraction and fusion enhancementof image information.

Description

technical field [0001] The embodiments of the present application relate to the technical field of target detection, and in particular, to a target detection method, system, computer device, and storage medium. Background technique [0002] In many computer vision application scenarios, such as object recognition, object detection is an important pre-step. Specifically, the region where the target is located is located by processing the target-related features in the collected images. [0003] The target may be a person, an object, or the like. Target detection models, such as neural network models, are used in target detection. [0004] With the rapid development of target detection model technology, there have been many improvements in accuracy and efficiency, but in complex scenarios in practical applications, there are still deficiencies. [0005] For example, text detection is a widely used type of target detection, and it is a pre-step for many computer vision tasks...

AI Technical Summary

Problems solved by technology

[0006] According to the technical characteristics of the methods used in text detection, the currently popular text detection methods can be roughly divided into two categories. The first category is the text detection method based on sliding windows. It is mainly based on the idea of ​​general target detection, setting a large number of different length and width. Ratio, anchor boxes of different sizes, use these anchor boxes as sliding windows, perform traversal search on the image or on the feature map obtained from the convolution operation on the image, for each searched position box, Classify and judge whether the frame is text. The advantage of this method is that after the text frame is judged, follow-up work can be carried out without the rest of the follow-up. The disadvantage is that the amount of calculation is too large, which not only consumes a lot of computing resources, but also takes time. Longer; the second type is based on the method of calculating the connected domain, also known as the method based on the idea of ​​​​segmentation. It mainly uses the full convolutional neural network model to extract image features, and then binarizes the feature map and calculates its connected domain. , and then according to different application scenarios (that is, different training data sets), use some specific methods to judge the position of the text line. The advantage of this method is that the calculation is fast and the calculation amount is small; however, the disadvantage is that the post-processing steps are cumbersome and involve a lot of The calculation and tuning of the algorithm will not only consume a lot of time, but also whether the post-processing strategy is reasonable and effective also strictly restricts the performance of the algorithm.

[0007] For the second method of segmentation ideas, there is a lot of research on how to simplify the post-processing part, but it is useful for text detection in scenes with less text, for example, in some open scene text detection data On the set (4 to 5 text boxes per image), the effect is acceptable, but for practical application scenarios with very dense text (such as hundreds of text areas on an image, such as elementary school students' arithmetic exercise books, etc.), the detection speed It is greatly affected by the number of text boxes, and almost linearly decreases with the increase of the number of text boxes, which cannot meet the speed requirements of actual application scenarios

Images