【论文阅读】3-Plan3D: Viewpoint and Trajectory Optimization for Aerial Multi-View Stereo Reconstruction

1、video

youtube

2、procedure

2.1、ROI

2.2、Initial set of images

2.3、Initial restruction (sfm\mvs)

2.4、Volumetric occupancy map(occupied/ free-space/unobserved voxels)

Value --: 0(empty) /0.5(unknowed)/ 1(knowed occupied)
Viewpoint-pixel-ray-caste-3D point(parallel projection)

2.5、Graph(node-viewpoint edge-path motion)

2.5.1、Viewpoint

1. Initial sample: candidate node(dense in ROI, sparse out ROI)
2. Camera pose—idealize—far away ,fronto-parallel,roll-0, pitch(-pi/2,0)
3. Maximize total information:（重点）

1）目标函数：（计算公式逐步细化）

incorporate stereo matching: （the edge selection strategy also includes） ----This encourages on average 1/ξ cameras to observe the same point:

Incidence: angle(incident viewing-surface normal)

Resolution: the number of pixels

2）优化方法:

Greedy （next viewpoint-far）/cost-benefit（next viewpoint-close）—折中
Recursive
split the budget into fisrt-middle & middle-end

4. Travel budget
5. Free-space\Collision freedom (according to the occupancy map)
6. Additional viewpoint: sparse matching

2.5.2、path

Bounding box(drone\obstacle)
RRT(rapidly exporing random trees)
Between the sparse matching

2.5.3、Edge wight

Real distance

2.6、Assess

Precision Recall F1 & F
Ground truth \ Roberts
Regular baseline pattern(circles,meanders,hemispheres)

2.7、Limiting:

• Possible viewpoints-finite (distance away obstacle)
• Split two scanning phases(change quickly environment- not suitable)
• Obstacle: ignore power line(initial restructure)