-
1.发明授权公开(公告)号：US09298988B2
公开(公告)日：2016-03-29
申请号：US14076030
申请日：2013-11-08
Applicant: ANALOG DEVICES TECHNOLOGY
Inventor： Prasanna Baja Thirumaleshwara , Vijaykumar Nagarajan
CPC classification number: G06K9/00624 , G06K9/4604 , G06K9/4642 , G06K9/6267 , G06K9/6269 , G06N99/005 , G06T11/60
Abstract: An exemplary object detection method includes generating feature block components representing an image frame, and analyzing the image frame using the feature block components. For each feature block row of the image frame, feature block components associated with the feature block row are evaluated to determine a partial vector dot product for detector windows that overlap a portion of the image frame including the feature block row, such that each detector window has an associated group of partial vector dot products. The method can include determining a vector dot product associated with each detector window based on the associated group of partial vector dot products, and classifying an image frame portion corresponding with each detector window as an object or non-object based on the vector dot product. Each feature block component can be moved from external memory to internal memory once implementing the exemplary object detection method.
Abstract translation: 示例性对象检测方法包括生成表示图像帧的特征块分量，以及使用特征块分量来分析图像帧。 对于图像帧的每个特征块行，评估与特征块行相关联的特征块分量，以确定与包括特征块行的图像帧的一部分重叠的检测器窗口的部分矢量点积，使得每个检测器窗口 具有相关组的部分矢量点积。 该方法可以包括基于相关组的部分矢量点积来确定与每个检测器窗口相关联的矢量点积，并且基于矢量点积将对应于每个检测器窗口的图像帧部分分类为对象或非对象。 一旦实现了示例性对象检测方法，每个特征块组件可以从外部存储器移动到内部存储器。
-
2.发明申请公开(公告)号：US20150131848A1
公开(公告)日：2015-05-14
申请号：US14076030
申请日：2013-11-08
Applicant: ANALOG DEVICES TECHNOLOGY
Inventor： Prasanna Baja Thirumaleshwara , Vijaykumar Nagarajan
CPC classification number: G06K9/00624 , G06K9/4604 , G06K9/4642 , G06K9/6267 , G06K9/6269 , G06N99/005 , G06T11/60
Abstract: An exemplary object detection method includes generating feature block components representing an image frame, and analyzing the image frame using the feature block components. For each feature block row of the image frame, feature block components associated with the feature block row are evaluated to determine a partial vector dot product for detector windows that overlap a portion of the image frame including the feature block row, such that each detector window has an associated group of partial vector dot products. The method can include determining a vector dot product associated with each detector window based on the associated group of partial vector dot products, and classifying an image frame portion corresponding with each detector window as an object or non-object based on the vector dot product. Each feature block component can be moved from external memory to internal memory once implementing the exemplary object detection method.
Abstract translation: 示例性对象检测方法包括生成表示图像帧的特征块分量，以及使用特征块分量来分析图像帧。 对于图像帧的每个特征块行，评估与特征块行相关联的特征块分量，以确定与包括特征块行的图像帧的一部分重叠的检测器窗口的部分矢量点积，使得每个检测器窗口 具有相关组的部分矢量点积。 该方法可以包括基于相关组的部分矢量点积来确定与每个检测器窗口相关联的矢量点积，并且基于矢量点积将对应于每个检测器窗口的图像帧部分分类为对象或非对象。 一旦实现了示例性对象检测方法，每个特征块组件可以从外部存储器移动到内部存储器。
-
Search Results
2
records
(took 0.01 seconds)
