Model-based play field registration
AssigneeIntel Corporation
InventorsXiaofeng Tong, Wenlong Li, Yimin Zhang
Life of this patent
Estimating the term…
Abstract
A method, apparatus, and system are described for model-based playfield registration. An input video image is processed. The processing of the video image includes extracting key points relating to the video image. Further, whether enough key points relating to the video image were extracted is determined, and a direct estimation of the video image is performed if enough key points have been extracted and then, a homograph matrix of a final video image based on the direct estimation is generated.
Claims16
A method comprising:
- processing an input video image, wherein processing including extracting key points relating to the video image;
- detecting whether enough key points relating to the video image were extracted, wherein the detecting includes searching for optimal mapping, wherein the searching includes calculating a matching error by using a distance map generated by a distance transform, and further includes one of (i) matching of angles of detected lines or a field position, (ii) a size of a field model, and (iii) comparing a camera parameter to a predetermined range;
- performing direct estimation of the video image if enough key points have been extracted;
- and
- generating a homograph matrix of a final video image based on the direct estimation.
A system having a storage medium and a processor coupled with the storage medium, the processor to:
- process an input video image, wherein processing including extracting key points relating to the video image;
- detect whether enough key points relating to the video image were extracted, wherein the detecting includes searching for optimal mapping, wherein the searching includes calculating a matching error by using a distance map generated by a distance transform, and further includes one of (i) matching of angles of detected lines or a field position, (ii) a size of a field model, and (iii) comparing a camera parameter to a predetermined range;
- perform direct estimation of the video image if enough key points have been extracted;
- and
- generate a homograph matrix of a final video image based on the direct estimation.
A non-transitory machine-readable medium comprising instructions which, when executed, cause a machine to:
- process an input video image, wherein processing including extracting key points relating to the video image;
- detect whether enough key points relating to the video image were extracted, wherein the detecting includes searching for optimal mapping, wherein the searching includes calculating a matching error by using a distance map generated by a distance transform, and further includes one of (i) matching of angles of detected lines or a field position, (ii) a size of a field model, and (iii) comparing a camera parameter to a predetermined range;
- perform direct estimation of the video image if enough key points have been extracted;
- and
- generate a homograph matrix of a final video image based on the direct estimation.
Dependencies are inferred from the claim text.