Granted Patent B2

US 12,651,360 · App. 18/142,445 · Granted Jun 9, 2026

Inventors: Wenwen Tu (San Jose, CA); Zuoguan Wang (Los Gatos, CA); Qun Gu (San Jose, CA)

Assignee: Black Sesame Technologies Inc.

G06T7/50G06T2207/10028G06T2207/20081

Abstract

Claims (34)

1 . A method for estimating monocular depth, comprising:

capturing an image;

encoding the image to form an encoded image;

comparing a projected mask in one or more dynamic objects with a pre-defined mask in the encoded image to predict an overlap ratio;

updating a pre-defined one or more dynamic objects as a static object or a dynamic object based on the overlap ratio;

masking one or more dynamic objects in the encoded image;

predicting a final prediction performance based on one or more unmasked static objects from the encoded image;

decoding and projecting the final prediction performance;

producing an information based on Global Positioning System (GPS), Inertial Measurement Unit (IMU), and wheel encoder fusion; and

estimating the monocular depth based on the final prediction performance and the information.

2 . A method for estimating monocular depth, comprising:

capturing an image;

encoding the image to form an encoded image;

comparing a projected mask in one or more dynamic objects with a pre-defined mask in the encoded image to predict an overlap ratio;

updating a pre-defined one or more dynamic objects to either a static object or a dynamic object based on the overlap ratio;

predicting a final prediction performance based on one or more unmasked static objects from the encoded image; and

decoding and projecting the final prediction performance;

producing an information based on Global Positioning System (GPS), Inertial Measurement Unit (IMU), and wheel encoder fusion; and

estimating the monocular depth based on the final prediction performance and the information.

3 . The method of claim 1 , wherein masking one or more dynamic objects in the encoded image comprises masking the one or more dynamic objects with a self-detected mask on calculating loss in training.

4 . The method of claim 3 , wherein the self-detected mask is formed by comparing an estimated depth and a projected depth map.

5 . The method of claim 1 , wherein the one or more dynamic objects is masked from a final loss function.

6 . The method of claim 5 , wherein the final loss function is calculated to improve the final prediction performance.

7 . The method of claim 6 , wherein the final loss function comprises a re-projection loss, a smoothness loss, and a geometry consistency loss.

8 . The method of claim 7 , wherein the re-projection loss is summation of photometric loss and structural similarity (SSIM) difference.

9 . The method of claim 7 , wherein the geometry consistency loss comprises one or more re-projection weights.

10 . A system for estimating monocular depth, comprising:

a memory and a processor, wherein the memory is connected to the processor;

the memory stores a computer program; and

the processor implements the method of claim 1 when executing the computer program.

11 . A system for estimating monocular depth, comprising:

a processor, wherein the memory is connected to the processor;

the memory stores a computer program; and

the processor implements the method of claim 2 when executing the computer program.

Assignments (2)

Continuity (1)

References Cited (35)