白丝美女被狂躁免费视频网站,500av导航大全精品,yw.193.cnc爆乳尤物未满,97se亚洲综合色区,аⅴ天堂中文在线网官网

Image processing apparatus, image processing system, and image processing method

專利號
US10867166B2
公開日期
2020-12-15
申請人
SONY CORPORATION(JP Tokyo)
發(fā)明人
Kazuhiro Shimauchi; Takashi Kohashi; Masakazu Ebihara; Yuji Ono; Hidetoshi Nagano
IPC分類
G06K9/00; G06T7/215; G06T7/292; G06K9/62; G06T7/20
技術(shù)領(lǐng)域
person,detection,candidate,in,image,frame,moving,body,process,point
地域: Tokyo

摘要

Provided are an apparatus and a method for executing a person detection process from an image. A divisional small region setting window is set and an image in the window is determined to be a person based on the number of moving body pixels of each divisional small region, and a first person candidate point is set in the window when the image is determined to be a person. A person determination process is executed based on an image feature quantity in a person determination detection frame, and a second person candidate point is set in the person determination detection frame when it is determined to be a person. A distribution state in a moving body detection frame of the person determination detection frame is analyzed, and an image in the person determination detection frame is determined to be a person based on an analysis result.

說明書

A specific example of the first person candidate point list will be described with reference to FIGS. 15 and 16.

FIG. 15 illustrates an example of the first person candidate point list.

Five entries, that is, identifiers 0 to 4 are registered in the first person candidate point list illustrated in FIG. 15.

The five pieces of registration data (entries) are registration data corresponding to the “first person candidate point” set for the moving body determined as a person obtained by executing the flow illustrated in FIG. 8.

Specifically, for example, it is registration data (entry) corresponding to each first person candidate point 161 illustrated in FIG. 12.

Further, one “first person candidate point” is detected corresponding to one setting position of the divisional small region setting window, and the “first person candidate point” is also set in a plurality of windows adjacent to one window in which the “first person candidate point” is set.

Both of them may be determined to be detection points corresponding to one person detected from the image, and the “first person candidate point” at the proximity positions may be merged into one piece of data and registered in a list.

Further, as a merging technique for merging a plurality of points at adjacent coordinate positions into one point, a merging technique known from the past is applied.

權(quán)利要求

1
微信群二維碼
意見反饋