Combined Edge- and Stixel-based Object Detection in 3D Point Cloud

. 2019 Oct 12;19(20):4423. doi: 10.3390/s19204423

Algorithm 1 Edge- and Stixel-Oriented 3D Point Cloud Segmentation

Capture the corresponding images $i m a g e_{l e f t}$ , $i m a g e_{r i g h t}$ from the stereo camera.
Detect the edges ${e_{l}^{1}, e_{l}^{2}, \dots, e_{l}^{n}, e_{r}^{1}, e_{r}^{2}, \dots, e_{r}^{m} | n, m \in N}$ using the structured forests-based edge detector.
Generate the edge candidate points ${c_{l}^{1}, c_{l}^{2}, \dots, c_{l}^{p}, c_{r}^{1}, c_{r}^{2}, \dots, c_{r}^{q} | p, q \in N}$ according to the rules given in the following Section 3.1.
Identify the utmost 4 bounding areas ${b_{l}^{1}, b_{l}^{2}, b_{l}^{3}, b_{l}^{4}, b_{r}^{1}, b_{r}^{2}, b_{r}^{3}, b_{r}^{4}}$ in each image based on the edge candidate points.
Generate the utmost 4 sparse point clouds (SPCs) ${S P C_{1}, S P C_{2}, S P C_{3}, S P C_{4}}$ of objects in the bounding areas.
Match the sparse point cloud $S P C_{i}$ with the dense point cloud $D P C$ .
Segment the dense point cloud $D P C_{j}$ of objects from the sensed dense point cloud $D P C$ .