Image/video target detection result enhancement method based on spatial-temporal tree filter

Abstract

The invention relates to an image/video target detection result enhancement method based on a spatial-temporal tree filter. The image/video target detection result enhancement method comprises the following steps: carrying out foreground fragment splicing and detection noise removal on a binary foreground detection image obtained by any video or image target detection algorithm; in the obtained detection image, carrying out spatial-temporal tree filter enhancement on the extracted foreground target area; and carrying out binarization operation on the filtering area. According to the image/video target detection result enhancement method, the global spatial-temporal tree filter is provided by utilizing the time continuity of the video sequence, and the detection image at the current momentcan be improved and enhanced by effectively utilizing all pixel contents of the same foreground target in a short-term history. Moreover, the image/video target detection result enhancement method combines the advantages of local filtering and global filtering, can achieve a better edge-preserving effect, and also considers the dense spatial-temporal correlation among all foreground pixels, thereby enabling the interior of an enhanced foreground detection target to be complete, enabling the contour to be clear, and remarkably reducing the noise.

Description

technical field [0001] The invention relates to the technical field of image or video target detection, in particular to a method for enhancing image / video target detection results based on spatio-temporal tree filters. Background technique [0002] The purpose of video object detection is to use algorithms to automatically detect areas of change or areas of interest in a series of images taken at different times of the same scene. [0003] Target detection algorithms often get noisy and under-detected results, and sometimes they are prone to over-smoothed targets. At present, no detection algorithm is suitable for all scenarios and task requirements. In order to obtain satisfactory detection results, there are some algorithms for enhancing the preliminary detection image. Commonly used algorithms include: image morphology operations, Markov random fields, and conditional random fields. The enhancement effect of image morphology operation is very limited, the commonly used...

AI Technical Summary

Problems solved by technology

However, image filtering has not been used in the field of video detection

Images