Target category correction method and detection method based on video dynamic foreground mask

Abstract

The invention belongs to the field of video image processing, particularly relates to a target category correction method and detection method based on a video dynamic foreground mask, and aims to solve the problem that target detection based on a convolutional neural network is poor in small target detection effect. The correction method comprises the following steps: acquiring a first correctionset of video frames; selecting a target rectangular frame greater than a confidence coefficient threshold as a candidate frame; taking pixel points greater than a gray threshold value in the foreground binary mask image as foreground points, for each candidate box, calculating the proportion of the foreground pixel points in the rectangular box and serving as a foreground score, fusing the scorewith the confidence coefficient of the maximum confidence target category of the candidate box to obtain the corrected confidence coefficient, and updating the maximum confidence target category confidence coefficient of the corresponding target object in the first correction set. According to the method, the target detection accuracy is improved, and the target detection effect that features ofsmall objects and the like are difficult to extract by the convolutional neural network can be particularly improved.

Description

technical field [0001] The invention belongs to the field of video image processing, and in particular relates to an object category correction method and a detection method based on a video dynamic foreground mask. Background technique [0002] Target detection is a technology that predicts the position of all target objects in an image, marks them with a rectangular frame, and predicts the category of the objects in the frame. At present, deep convolutional neural networks are often used to extract robust and expressive object features for target detection. Compared with the previous target detection method based on manual feature combination, the speed and accuracy of target detection are improved. [0003] Object detection based on convolutional neural network is divided into one-stage and two-stage. The one-stage detector inputs the image into an end-to-end basic neural network, and the network outputs the final rectangular frame position and category of the target obj...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the above-mentioned problems in the prior art, that is, in order to solve the problem that the target detection based on the neural network is poor for small target detection, the first aspect of the present invention proposes a target category correction based on the video dynamic foreground mask method, which includes:

Images