Playfield detection and shot classification in sports video

1. A method of classifying the shot type of a video frame, comprising:

loading a frame comprising a plurality of pixels into memory;

dividing said frame into field pixels and non-field pixels based on a first playfield detection criteria, wherein said dividing of said frame into said field pixels and said non-field pixels generates a binary mask;

determining an initial shot type classification using the number of said field pixels and the number of said non-field pixels;

partitioning said frame into one or more regions based on said initial classification;

determining the status of each of said one or more regions based upon the number of said field pixels and said non-field pixels located within each said region;

determining a shot type classification for said frame based upon said status of each said region;

determining a camera view angle from said binary mask;

generating a seed mask;

generating a local histogram of the mask pixels of said seed mask;

accumulating at least a predetermined number of said local histograms into an accumulated histogram for each camera view; and

identifying one or more color ranges for the key area of a basketball court from said accumulated histogram for one or more particular camera views.

2. The method of claim 1 , further comprising:

updating said first playfield detection criteria when said shot type classification for said frame is a long shot or a medium shot by generating one or more local histograms of the color components of said frame within a selected region of said frame;

generating one or more accumulated histograms for a final shot type classification by integrating data in said local histograms of the current frame and previous frames of said final shot type classification; and

updating said first playfield detection criteria for subsequent frames by narrowing said first playfield detection criteria to fit peaks on said accumulated histograms for the final shot type classification.

3. The method of claim 2 , wherein said color components are a hue component, a saturation component, and a value component.

4. The method of claim 2 , wherein said selection region is the bottom three quarters of said frame when said frame is divided vertically into two sections.

5. The method of claim 2 , wherein the step of updating the first playfield detection criteria further comprises:

calculating a value component threshold for the accumulated histogram for a final shot type classification;

dividing the value component of the accumulated histogram for a final shot type classification into a first region comprising value components less than the value component threshold and a second region comprising value components greater than the value component threshold; and

setting the value component of the first playfield detection criteria to include the value component values within the second region.

6. The method of claim 5 , wherein said value component threshold is the value component obtained by applying Otsu's method on the value components of the accumulated histogram for a final shot type classification to obtain a upper Otsu region and applying Otsu's method on the upper Otsu region to obtain the value component threshold.

7. The method of claim 5 , wherein said value component threshold is the value component obtained by:

finding a first threshold such that the distribution of value components in the accumulated histogram for a final shot type classification between 0 and the first threshold is not less than a first distribution threshold;

finding a second threshold such that the distribution of value components in the accumulated histogram for a final shot type classification between 0 and the second threshold is not less than a second distribution threshold; and

finding the minimum value component threshold value between the first distribution threshold and the second distribution threshold.

8. The method of claim 1 , wherein said first playfield detection criteria are one or more color ranges, wherein each said color range consists of a range of hue component values, a range of saturation component values, and a range of value component values.

9. The method of claim 1 , further comprising:

applying a second playfield detection criteria to a selected region of said frame to obtain field pixels for the selected region of said frame, wherein the color range of the second playfield detection criteria is different than the color range of the first playfield detection criteria;

incorporating the field pixels for the selected region of said frame into the field pixels obtained from the first playfield detection criteria.

10. The method of claim 1 , wherein one of said at least one regions is the bottom three quarters of said frame when said frame is divided vertically into two sections.

11. The method of claim 1 , wherein one of said at least one regions is the center half of said frame when said frame is divided horizontally into three sections.

12. The method of claim 1 , wherein said binary mask is generated from a logical combination of multiple masks each generated by a division of said frame into said field pixels and said non-field pixels according to different sets of color ranges in said first playfield detection criteria.

13. The method of claim 1 , wherein the step of determining a shot type classification for said frame further comprises:

calculating a standard deviation of the hue of the field pixels and the standard deviation of the hue of the non-field pixels;

comparing a standard deviation ratio by dividing the standard deviation of the non-field pixels by the standard deviation of the field pixels; and

comparing the standard deviation ratio to a predetermined standard deviation ratio threshold.