一、点云处理任务
英文名称 | 中文名称 | 说明 |
---|---|---|
Object classification | 物体分类 | 识别物体的类别名称 |
Parts segmentation | 部件分割 | 对一个物体的不同部件进行分割识别 |
Object detection | 目标检测 | 对点云表示的物体进行识别,并把感兴趣的物体用立体框框出 |
Semantic segmentation | 场景与语义分割 | 对场景中的物品进行分类 |
二、三维图像表示方法
名称 | 图片 |
---|---|
Point Cloud | ![]() |
Mesh | ![]() |
Volumetric | ![]() |
Projected View RGB D | ![]() |
三、点云数据的应用
- Robot Perception
- Augmented Reality
- Shape Design
- VR
- FaceID
四、训练数据集
- Datasets for 3D Shape Classification
- Datasets for 3D Object Detection and tracking
- Datasets for 3D Point Cloud Segmentation
五、挑战
- Irregular. Sparse and dense regions(激光雷达:进密远疏)
- Unstructured. No grid, each point is independent and distance between neighboring points is not fixed.
- unorderd.
导致置换不变性
六、处理方法
- Voxel based Methods
- multiview based Methods
- point-based Methods
a. PointNet
b. PointNet++ - Convolution-based Methods
a. 3D neighboring points
b. 3D continuous convolution
c. 3D discrete convolution - Graph-based Methods