Method for estimating the location of an image target region from tracked multiple image landmark regions

In an image processing method suitable for use in pattern-key insertion of extraneous foreground image data in a target region of a background image to derive thereby a composite image, said target region having a particular location with respect to a scene being viewed by an image sensor over a period of time, wherein said method employs a world map having stored therein the relative position of the location and the pose of at least one of multiple pre-trained reference image patterns of landmark regions in said scene with respect to that of said target region;

• wherein said method comprises computation steps for inferring the size and position of said particular location within each of ongoing successive image frames of said scene from the shape, size and position of said one of said multiple landmark regions represented within each of successive image frames of said scene;
• and wherein the intensity structure of said one of said multiple landmark regions represented within each of successive image frames of said scene may change over time with respect to the intensity structure of the pre-trained reference image pattern of said one of said multiple landmark regions;
• the improvement wherein said computation steps comprise the steps of:
• a) only initially employing a model whose image-change-in-position parameters are computed between the first-occurring image field of said successive image frames and the pre-trained reference image pattern for determining the shape, size and position of said one of said multiple landmark regions represented by the first-occurring image field of said successive image frames;
• and
• b) thereafter employing a model whose image-change-in-position parameters are dynamically computed in accordance with a given function of the number of those image fields of said successive image frames that precede the current image field for determining the shape, size and position of said one of said multiple landmark regions represented by the current image field of said successive image frames, wherein the position of said current image field of said one of said multiple landmark regions is P and said given function comprises the following equations:
• 0<T.ltoreq.T.sub.1, P=R(n.sub.0),
• and
• kT.sub.1 <T.ltoreq.(k+1)T.sub.1, P=R(n.sub.k),
• where T is the elapsed time since the beginning of the first-occurring image field of said successive image frames;
• T.sub.1 is a specified update time interval;
• k is an integer having a value of at least one;
• R,(n.sub.0) is a component of said model whose image-change-in-position parameters are computed between the current field image and the pre-trained reference image pattern, and R(n.sub.k) is a component of said model whose image-change-in-position parameters are computed between the presently current field image and that field image which was current at time kT.sub.1.

In an image processing method suitable for use in pattern-key insertion of extraneous foreground image data in a target region of a background image to derive thereby a composite image, said target region having a particular location with respect to a scene being viewed by an image sensor over a period of time, wherein said method employs a world map having stored therein the relative position of the location and the pose of at least one of multiple pre-trained reference image patterns of landmark regions in said scene with respect to that of said target region;

• wherein said method comprises computation steps for inferring the size and position of said particular location within each of on going successive image frames of said scene from the shape, size and position of said one of said multiple landmark regions represented within each of successive image frames of said scene;
• and wherein the intensity structure of said one of said multiple landmark regions represented within each of successive image frames of said scene may change over time with respect to the intensity structure of the pre-trained reference image pattern of said one of said multiple landmark regions;
• the improvement wherein said computation steps comprise the steps of:
• a) only initially employing a model whose image-change-in-position parameters are computed between the first-occurring image field of said successive image frames and the pre-trained reference image pattern for determining the shape, size and position of said one of said multiple landmark regions represented by the first-occurring image field of said successive image frames;
• and
• b) thereafter employing a model whose image-change-in-position parameters are dynamically computed in accordance with a given function of the number of those image fields of said successive image frames that precede the current image field for determining the shape, size and position of said one of said multiple landmark regions represented by the current image field of said successive image frames, wherein the position of said current image field of said one of said multiple landmark regions is P;
• and said given function comprises the equations
• 0<T.ltoreq.T.sub.1, P=R(n.sub.0),
• and
• T>T.sub.1 P=.alpha.*R(n.sub.k)+(1-.alpha.)*R(n.sub.0),
• where T is the elapsed time since the beginning of the first-occurring image field of said successive image frames;
• T.sub.1 is a specified update time interval;
• k is an integer having a value of at least one;
• R(n.sub.0) is a component of said model whose image-change-in-position parameters are computed between the current field image and the pre-trained reference image pattern;
• R(n.sub.k) is a component of said model whose image-change-in-position parameters are computed between the presently current field image and that field image which was current at time kT.sub.1, and where .alpha. is a weighting parameter having a value larger than 0 and smaller than 1.