Methods and systems for detecting and locating skipped frames in a video sequence
View Patent ↗Embodiments of the present invention comprise methods and systems for detecting and locating skipped frames in a test video sequence in relation to a reference video sequence. Some embodiments comprise identifying pairs of temporally aligned reference and test segments. An alignment offset and freeze-frame count associated with each segment pair is received and used to calculate a number of skipped frames between a first segment pair and a second segment pair, the second segment pair being temporally subsequent to the first segment pair. In some embodiments, a number of skipped frames between the first segment pair and the second segment pair is determined by calculating a segment offset between the two segment pairs and subtracting the segment offset value from the sum of the number of freeze frames associated with the first segment pair and all temporally intervening segment pairs between the first segment pair and the second segment pair.
1. A method for detecting skipped frames between a test video sequence and a reference video sequence, said method comprising:
a) identifying a first plurality of segment pairs associated with a test video sequence comprising a plurality of test frames and a reference video sequence comprising a plurality of reference frames, wherein each segment pair in the first plurality of segment pairs comprises a portion of the reference frames and a temporally aligned portion of the test frames;
b) for each segment pair in the first plurality of segment pairs, receiving an associated alignment offset and an associated freeze-frame count;
c) calculating a segment alignment offset between a first segment pair in the first plurality of segment pairs and a second segment pair in the first plurality of segment pairs, wherein the second segment pair is temporally subsequent to the first segment pair;
d) for the first segment pair and each segment pair in a second plurality of segment pairs comprising any segment pair in the first plurality of segment pairs temporally intervening between the first segment pair and the second segment pair, summing the associated freeze-frame counts; and
e) subtracting the segment alignment offset from the summed freeze-frame count to form a skipped-frame count associated with the first segment pair and the second segment pair.
2. The method as described in claim 1 further comprising:
a) for the first segment pair, calculating a similarity measure between corresponding frames in the first-segment-pair reference frames and the first-segment-pair temporally aligned test frames, thereby producing a first plurality of similarity measures;
b) for each segment pair in the second plurality of segment pairs, calculating a similarity measure between corresponding frames in the segment-pair reference frames and the segment-pair temporally aligned test frames, thereby producing a second plurality of similarity measures;
c) identifying discontinuities in said first plurality of similarity measures and said second plurality of similarity measures; and
d) resolving said identified discontinuities using the skipped-frame count.
3. The method as described in claim 2 further comprising spatially aligning corresponding frames in the first-segment-pair reference frames and the first-segment-pair temporally aligned test frames.
4. The method as described in claim 2 , wherein the calculating a similarity measure between corresponding frames in the first-segment-pair reference frames and the first-segment-pair temporally aligned test frames comprises calculating a measure selected from the group consisting of an error measure and a correlation measure.
5. The method as described in claim 1 , wherein the identifying a first plurality of segment pairs comprises:
a) distilling each test frame in the plurality of test frames and each reference frame in the plurality of reference frames into frame distillation measurements;
b) from the frame distillation measurements, performing a linear alignment measurement using a linear Hough transform of a local Pearson's cross-correlation coefficient (LPCCC) image for spatial alignment of the frames between the test and reference video sequences to find a best fit line through the LPCCC image; and
c) for each pixel of the LPCCC image along the best fit line, searching vertically for a higher correlation coefficient when the pixel has a value less than a threshold to find a better frame match for temporal alignment of the frames between the test and reference video sequences.
6. The method as described in claim 5 , wherein the frame distillation measurements comprise summaries of mean, standard deviation and root-mean-square (RMS) differences between corresponding frames of the test video sequence and the reference video sequence.
7. The method as described in claim 5 , wherein the frame distillation measurements comprise summaries of mean, standard deviation and root-mean-square (RMS) differences between sub-images for each pair of corresponding frames of the test video sequence and the reference video sequence.
8. The method as described in claim 5 , wherein the frame distillation measurements comprise summary line and summary column one-dimensional images of the LPCCC image.