Granted Patent B2

US 12,445,612 · App. 18/296,797 · Granted Oct 14, 2025

Inventors: Eldhose K. George (Kerala, IN); Jitendra S. Chaurasia (Bengaluru, IN); Indranil Datta (Bangalore, IN)

Assignee: HONEYWELL INTERNATIONAL INC.

H04N19/124H04N19/172H04N19/177H04N19/60

Abstract

Claims (62)

1. A method for compressing a video stream, comprising:

retrieving a plurality of frames corresponding to the video stream;

for each of two or more sequential frames of the plurality of frames of the video stream:

extracting Key Point Descriptors (KPDs) for the respective frame;

processing the respective frame using Principle Component Analysis (PCA) followed by vector quantization, resulting in a quantized explained variance matrix for the respective frame;

storing the quantized explained variance matrix for the respective frame, the quantized explained variance matrix represents the respective frame with reduced dimensions; and

storing the KPDs for the respective frame.

2. The method of claim 1 , comprising decoding each of the two or more sequential frames of the plurality of frames by:

for each of two or more sequential frames:

retrieving the quantized explained variance matrix for the respective frame and the KPDs for the respective frame;

performing an inverse PCA transform to the quantized explained variance matrix of the respective frame to produce a respective reconstructed frame;

merging the KPDs associated with the respective frame with the respective reconstructed frame; and

assembling the reconstructed frames for the two or more sequential frames into a decoded video sequence.

3. The method of claim 1 , wherein the respective frame is processed using Principle Component Analysis (PCA) to produce an intermediate explained variance matrix, followed by processing the intermediate explained variance matrix using vector quantization to produce the quantized explained variance matrix for the frame.

4. The method of claim 1 , wherein the two or more sequential frames comprise a Group of Pictures (GOP) that includes an I-Frame and one or more P-Frames.

5. The method of claim 4 , comprising extracting the Key Point Descriptors (KPDs) for each of the I-Frame and each of the P-Frames of the GOP.

6. The method of claim 5 , comprising processing each of the I-Frame and each of the P-Frames of the GOP using Principle Component Analysis (PCA) followed by vector quantization, resulting in a quantized explained variance matrix for the respective frame.

7. The method of claim 1 , wherein each of the plurality of frames corresponds to a mean subtracted frame (MSF), wherein the mean subtracted frame (MSF) represents an original frame of the video stream with a mean value subtracted from each pixel of the original frame.

8. A method for compressing a video stream, comprising:

receiving a video stream having a plurality of frames, wherein two or more sequential frames of the plurality of frames are associated with a Group of Pictures (GOP) that includes an I-Frame and one or more P-Frames;

extracting Key Point Descriptors (KPDs) for each of the two or more sequential frames of the GOP;

determining a mean frame for the GOP from the two or more sequential frames of the GOP;

subtracting from each of the two or more sequential frames of the GOP the mean frame, resulting in two or more mean subtracted frames (MSF) for the GOP;

for each of the mean subtracted frames (MSF) of the GOP:

processing the respective frame using Principle Component Analysis (PCA) followed by vector quantization, resulting in a quantized explained variance matrix for the respective frame, wherein the quantized explained variance matrix represents the respective frame with reduced dimensions;

storing the quantized explained variance matrix for the respective frame; and

storing the KPDs for the respective frame.

9. The method of claim 8 comprising decoding each of the two or more sequential frames of the GOP by:

retrieving the quantized explained variance matrix for the respective frame and the KPDs for the respective frame;

performing an inverse PCA transform to the quantized explained variance matrix of the respective frame to produce a respective reconstructed frame;

adding the respective reconstructed frame to the mean frame of the GOP to result in a reconstructed original frame;

merging the KPDs associated with the respective frame with the respective reconstructed original frame; and

assembly the reconstructed original frames into a decoded video sequence.

10. The method of claim 9 , comprising transmitting the quantized explained variance matrix and the KPDs for the respective frame to a storage device over a network.

11. The method of claim 10 , comprising retrieving the quantized explained variance matrix for the respective frame and the KPDs for the respective frame from the storage device to decode the respective frame.

12. A method for compressing a video stream, comprising:

receiving a plurality of frames corresponding to the video stream, wherein two or more sequential frames of the plurality of frames are associated with a Group of Pictures (GOP) that includes an I-Frame and one or more P-Frames;

extracting Key Point Descriptors (KPDs) for each of the two or more sequential frames of the GOP;

storing the KPDs for each of the two or more sequential frames of the GOP;

determining a mean frame for the GOP from the two or more sequential frames of the GOP;

storing the mean frame;

subtracting each of the two or more sequential frames of the GOP from the mean frame, resulting in two or more mean subtracted frames (MSF) for the GOP;

for each of the mean subtracted frames of the GOP:

dividing the respective frame into a plurality of blocks of pixels;

for each of the blocks:

processing the respective block using Principle Component Analysis (PCA) followed by vector quantization, resulting in a quantized explained variance matrix for the respective block, wherein the quantized explained variance matrix represents the respective block with reduced dimensions; and

storing the quantized explained variance matrix for the respective block.

13. The method of claim 12 , comprising transmitting the quantized explained variance matrix for each of the respective blocks of the respective frame to a storage device over a network.

14. The method of claim 12 , comprising sequentially selecting each of the plurality of blocks in a selection order, and:

processing the respective block using Principle Component Analysis (PCA) followed by vector quantization, resulting in in a quantized explained variance matrix for the respective block; and

storing the quantized explained variance matrix for the respective block, the quantized explained variance matrix represents the respective block with reduced dimensions.

15. The method of claim 13 , wherein the selection order is sequential, random or pseudo-random.

16. The method of claim 12 comprising decoding each of the two or more sequential frames of the GOP by:

retrieving the quantized explained variance matrix for each of the blocks of the respective frame;

performing an inverse PCA transform to the quantized explained variance matrix of each of the blocks of the respective frame;

merging the transformed quantized explained variance matrices of the blocks of the respective frame to reconstruct the respective frame;

adding the mean frame to the respective reconstructed frame to result in a reconstructed original frame;

merging the KPDs associated with the respective frame with the respective reconstructed original frame; and

assembling the reconstructed original frames into a decoded video sequence.

17. The method of claim 12 , comprising:

receiving a dimension variable applicable to the GOP, wherein the Principle Component Analysis (PCA) retains a retained number of principle components for each of the respective blocks based at least in part on the dimension variable.

18. The method of claim 17 , wherein the dimension variable is updated by a video compression controller to control an amount of compression that is desired for the GOP.

Assignments (1)

Continuity (1)

References Cited (44)