首页> 外国专利> BEHAVIOR IDENTIFICATION METHOD, DEVICE AND COMPUTER-READABLE STORAGE MEDIUM

BEHAVIOR IDENTIFICATION METHOD, DEVICE AND COMPUTER-READABLE STORAGE MEDIUM

机译:行为识别方法,设备和计算机可读存储介质

摘要

The present solution relates to artificial intelligence. Provided are a behavior identification method, a device and a storage medium. Said method comprises: segmenting a video stream into an image frame sequence; detecting a human body contour in each frame, and marking each human body with a first rectangular box; calculating the distance between any two first rectangular boxes in each frame; if the distance between two first rectangular boxes in a certain frame is less than a threshold, using a two-person combination box to enclose the two first rectangular boxes; searching a plurality of preceding and following frames, enclosing two persons which are the same as those in said two-person combination box to form a two-person combination box, and forming a two-person combination box sequence by using the two-person combination boxes in said frame and in the plurality of preceding and following frames; and inputting the two-person combination box sequence into a neural network model to perform behavior identification. The present application avoids a large amount of calculation of the neural network model caused by the unnecessary background, and also reserves features which are valuable for behavior determination between two persons, thereby improving the accuracy of behavior identification in a complex scenario.
机译:本解决方案涉及人工智能。提供了一种行为识别方法,设备和存储介质。所述方法包括:将视频流分段为图像帧序列;在每个框架中检测人体轮廓,并用第一矩形盒标记每个人体;计算每个框架中任意两个第一矩形盒之间的距离;如果某个框架中的两个第一矩形框之间的距离小于阈值,则使用双人组合框包围两个第一矩形盒;在多个之前的帧中搜索多个和以下帧,包含与所述双人组合框中的两个人相同,以形成双人组合框,并通过使用双人组合形成双人组合框序列在所述框架和多个前后框架中的框;并将双人组合框序列输入到神经网络模型中以执行行为识别。本申请避免了由不必要的背景引起的神经网络模型的大量计算,并且还储存对两个人之间的行为确定有价值的特征,从而提高了复杂场景中的行为识别的准确性。

著录项

相似文献

  • 专利
  • 外文文献
  • 中文文献
获取专利

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号