首先试试这些参数：
feature_min_frames: 1000 # If fewer frames are detected, sift_peak_threshold/surf_hessian_threshold is reduced.
processes: 4 # Number of threads to use
feature_type: LF
feature_root: 1
LFnetFeaturePath: ./rgb_feats_indoor/
# lowes_ratio: 0.9
feature_process_size: -1
feature_root: 0
matching_gps_distance: 0 # Maximum gps distance between two images for matching
matching_gps_neighbors: 0 # Number of images to match selected by GPS distance. Set to 0 to use no limit (or disable if matching_gps_distance is also 0)
matching_time_neighbors: 0 # Number of images to match selected by time taken. Set to 0 to disable
matching_order_neighbors: 2 # Number of images to match selected by image name. Set to 0 to disable
preemptive_max: 500 # Number of features to use for preemptive matching
lowes_ratio: 0.6
bundle_interval: 999999
local_bundle_radius: 1362
主要思路是不希望用局部的BA优化而是用一次全局的BA
之前尝试局部做效果比较差
运行下流程看看结果：
2019-06-02 11-11-28屏幕截图.png
2019-06-02 11-12-38屏幕截图.png
2019-06-02 11-12-21屏幕截图.png
实验结果比较垃圾．．．．．．
再试一下参数：
feature_min_frames: 1000
processes: 4
feature_type: LF
feature_root: 1
LFnetFeaturePath: ./rgb_feats_indoor/
# lowes_ratio: 0.9
feature_process_size: -1
feature_root: 0
matching_gps_distance: 0
matching_gps_neighbors: 0
matching_time_neighbors: 0
matching_order_neighbors: 100
preemptive_max: 500
lowes_ratio: 0.8
bundle_interval: 999999
local_bundle_radius: 1362
上次有可能使临近帧match太少(只有两帧，类似video模式)导致tracks太少不够用
这次改了１００帧，肯定是够用了．
结果如下：
稍微好了一点把，起码是把图片串到一起了
再试试新的参数：
feature_min_frames: 1000
processes: 8
feature_type: LF
feature_root: 1
LFnetFeaturePath: ./rgb_feats_indoor/
# lowes_ratio: 0.9
feature_process_size: -1
feature_root: 0
matching_gps_distance: 0
matching_gps_neighbors: 0
matching_time_neighbors: 0
matching_order_neighbors: 100
preemptive_max: 500
lowes_ratio: 0.8
bundle_interval: 999999
local_bundle_radius: 100
2019-06-02 13-42-21屏幕截图.png
