Stereo vision concept in Computer Vision - Model Pipeline Trace

Model Pipeline - Stereo vision concept

Stereo vision uses two images from slightly different viewpoints to understand depth, like how our eyes see the world in 3D. It finds matching points in both images to calculate how far objects are.

Data Flow - 5 Stages

1Capture stereo images

2 images of 480 x 640 pixels each→Take two photos from cameras placed side-by-side→2 images of 480 x 640 pixels each

Left image and right image of a room from slightly different angles

↓

2Preprocessing

2 images of 480 x 640 pixels each→Convert images to grayscale and normalize brightness→2 grayscale images of 480 x 640 pixels each

Left and right grayscale images with pixel values between 0 and 1

↓

3Feature matching

2 grayscale images of 480 x 640 pixels each→Find matching points between left and right images→List of matched points with coordinates in both images

Point (150, 200) in left image matches point (140, 200) in right image

↓

4Disparity calculation

List of matched points→Calculate horizontal pixel difference (disparity) between matched points→Disparity map of size 480 x 640 pixels

Disparity value 10 pixels at location (150, 200)

↓

5Depth estimation

Disparity map of 480 x 640 pixels→Convert disparity values to depth using camera parameters→Depth map of 480 x 640 pixels with distance values

Depth value 2.5 meters at pixel (150, 200)

Training Trace - Epoch by Epoch

Loss
0.5 |****
0.4 |*** 
0.3 |**  
0.2 |*   
0.1 |*   
     1 2 3 4 5 Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	0.45	0.60	Model starts learning to match points between images
2	0.30	0.75	Matching accuracy improves, loss decreases
3	0.20	0.85	Model better at finding correct matches
4	0.15	0.90	Loss continues to decrease, accuracy nears 90%
5	0.12	0.92	Model converges with good matching performance

Prediction Trace - 4 Layers

Layer 1: Input stereo images

Layer 2: Feature matching

Layer 3: Disparity calculation

Layer 4: Depth estimation

Model Quiz - 3 Questions

What does the disparity value represent in stereo vision?

A The brightness difference between images

B The horizontal pixel difference between matched points

C The vertical pixel difference between matched points

D The color difference between images

Key Insight

Stereo vision uses differences between two images to estimate depth. Training improves the model's ability to find matching points, reducing error and increasing accuracy, which leads to better depth maps.