Image processing method and device for detecting object in video and image processing equipment

Abstract

The embodiment of the invention provides an image processing method and device for detecting an object in a video and image processing equipment. The method aims at a current frame image to be processed in the video, comprising the steps of extracting an initial feature map of the current frame image; obtaining a first reconstruction feature map and a second reconstruction feature map by correlation matching based on the initial feature map and a historical feature map of the current frame image; and performing feature fusion on the initial feature map, the first reconstruction feature map andthe second reconstruction feature map to obtain a fused feature map. Therefore, the image processing method of the embodiment of the invention is improved based on the initial feature map and the historical feature map, avoids the post-processing of the engineering skill, and can ensure the relative integrity of the object detection frame, can realize the real-time video object detection, therebyeffectively improving the detection effect and ensuring the detection accuracy.

Description

technical field [0001] The present invention relates to the field of image processing, and more particularly relates to an image processing method, device and equipment for object detection in video. Background technique [0002] Object detection algorithms based on deep learning, such as Region-based Convolutional Neural Network (R-CNN), are mostly designed and implemented for static images, but object detection in real-world applications is mostly used in videos For example, automatic driving, intelligent security, etc., and there will be fast movement of objects, camera movement or shaking in the video, which will cause the detection effect of some frames to deteriorate. Doing non-maximum suppression (NMS) along the time series (Non-maximum suppression, NMS), pipeline convolutional neural network method (TubeletsConvolutional Neural Networks), etc. are some post-processing algorithms for object detection in videos, but they are engineering skills There is a lot of proces...

AI Technical Summary

Problems solved by technology

[0002] Object detection algorithms based on deep learning, such as Region-based Convolutional Neural Network (R-CNN), are mostly designed and implemented for static images, but object detection in real-world applications is mostly used in videos For example, automatic driving, intelligent security, etc., and there will be fast movement of objects, camera movement or shaking in the video, which will cause the detection effect of some frames to deteriorate

Doing non-maximum suppression (NMS) along the time series (Non-maximum suppression, NMS), pipeline convolutional neural network method (TubeletsConvolutional Neural Networks), etc. are some post-processing algorithms for object detection in videos, but they are engineering skills There are many processes, and they are mostly used for post-processing of detection results, which are difficult to reproduce and cannot be applied in real-time detection environments

Moreover, these algorithms use the optical flow method to fuse the feature layer in the feature extraction stage, resulting in the error caused by the calculation will be accumulated in the feature fusion stage, which in turn affects the accuracy of the entire detection

Images