Appendix A: Algorithm Principles Used

This section introduces the core algorithm principles used in the project.

1. Stereo Matching

Generate a disparity map by calculating the disparity of the left and right images.

Core Steps

1. Block Matching: Calculate the similarity of the left and right image blocks.

2. Disparity Optimization: Smooth the disparity map through the WLS filter to reduce noise.

---

2. V-Disparity Ground Detection

Detect the ground by constructing a v-disparity map.

Core Steps

1. Count the frequency distribution of each column of disparity values.

2. Fit a straight line on the v-disparity map to represent the ground.

---

3. Stixel Technology

Stixel is a compressed scene representation method that summarizes each column of pixels into a columnar pixel.

Core steps

1. Analyze the disparity map by column and extract the ground pixels and top pixels of the object in each column.

2. Use the depth information to estimate the height and distance of the object in each column.

---

4. Stixel segmentation

Clustering method based on depth (Z direction) and column position (X direction).

Core steps

1. Z direction clustering: Group stixels according to depth.

2. X direction clustering: Further separate adjacent targets.