0_1. 分析数据：

0_2. 步骤：

单帧：

1.提取地面，提取特征

2.单帧与静态帧匹配，

3.分别按匹配结果与IMU结果，计算单帧点云与静止帧中最近点距离，统计在不同距离范围中点的占比。

4.按照距离分配intensity

建图：

1.使用里程计位姿去畸变;

2.里程计位姿叠帧建图;

3.IMU位姿去畸变；

4.IMU叠帧建图；

5.与静止场景比较特征重合程度；

6.计算静止点云与地图中最近的点的距离，统计在不同距离范围中点的占比。

7.按照距离分配intensity

1. 单帧ICP(scan-to-scan)

第0帧（静止帧）与第251帧（动态帧，已去畸变）对比

1.1 位置/姿态变换结果对比

	IMU测量	里程计结果	所有点匹配	去地面点匹配
pos.x	11.9864419965346	12.113	12.2315209393	12.19829592444
pos.y	0.35545271630706	0.279951	0.20742369269	0.201242980913
pos.z	-0.027396272823	0.229392	0.03955666560	0.150457801652
roll	0.12564565735257	0.1990477499999	0.26103444641	0.075069163511
pitch	-0.01561664334409	0.0350390808002	-0.38558692648	-0.20547417157
yaw	1.713945651730975	1.7661607749408	1.92228541454	1.942115501283

1.2 使用不同特征匹配结果：

结论：
地面点可以有效约束z轴漂移。
去除地面点匹配的精度更高。
单独的柱面/立面等特征无法完全约束三个方向的运动。

原始点云第0帧与第251帧	惯导测量结果	LiDAR里程计测量结果
	TRANSLATION: 11.98644199653465 0.35545271630706643 -0.027396272823478673 ROTATION: 0.12564565735257952 -0.015616643344094927 1.7139456517309755	TRANSLATION: 12.113 0.279951 0.229392 ROTATION: 0.19904774999987535 0.03503908080022266 1.766160774940752

采用特征	匹配效果	匹配结果
所有点云		TRANSLATION: 12.231520939336493 0.2074236926917551 0.03955666560427022 ROTATION: 0.26103444641217954 -0.38558692648123116 1.9222854145442057
nonground		TRANSLATION: 12.198295924444954 0.20124298091307577 0.1504578016521565 ROTATION: 0.07506916351143701 -0.20547417157562423 1.9421155012833935
ground		TRANSLATION: -0.45032801957550905 -1.539946635341638 -0.04409875650376381 ROTATION: 0.061396539662778 0.0393528844241862 2.687272808286958
facade		TRANSLATION: 0.18602666396360037 0.003922289244317075 -0.659840212760906 ROTATION: 0.9523135428663062 -0.7056622717456797 2.622647650589112
pillar		TRANSLATION: 1.9201082213595009 1.6873773196469384 -0.018532779692659894 ROTATION: -1.5216757902856088 -0.8649583504104713 -0.7739914823982094
beam		TRANSLATION: -0.2902069118127943 1.1793059954475964 -0.25265109924299545 ROTATION: -2.5767941530217193 0.09564350376227218 0.563153967273353
vertex		TRANSLATION: 1.4010472774746574 -1.084962950016715 -0.733484141649384 ROTATION: 0.47538640627435874 -1.475592489579815 4.088730099925485

1.3 单帧点云与静止帧中点距分析：

实验方法：

分别按匹配结果与IMU结果，计算单帧点云与静止帧中最近点距离，统计在不同距离范围中点的占比。以此来反映匹配精度。

结论：

去除地面点后进行匹配，整体点云，以及非地面点云重合程度更好。
里程计得到的变换误差大于0.5m的点占比最少；地面点云重合程度相对更高。

	IMU测量矩阵	里程计测量矩阵	所有点ICP	非地面点ICP
全部点云 81836 points	dist < 0.1: 9463 (11.56%) [0.1, 0.2): 20743 (25.35%) [0.2, 0.3): 28139 (34.38%) [0.3, 0.4): 8918 (10.90%) [0.4, 0.5): 4516 (5.52%) [0.5, ~ ~): 10057 (12.29%)	dist < 0.1: 26723 (32.65%) [0.1, 0.2): 28379 (34.68%) [0.2, 0.3): 9651 (11.79%) [0.3, 0.4): 5171 (6.32%) [0.4, 0.5): 3300 (4.03%) [0.5, ~ ~): 8612 (10.52%)	dist < 0.1: 32270 (39.43%) [0.1, 0.2): 20806 (25.42%) [0.2, 0.3): 9305 (11.37%) [0.3, 0.4): 5685 (6.95%) [0.4, 0.5): 3771 (4.61%) [0.5, ~ ~): 9999 (12.22%)	dist < 0.1: 32990 (40.31%) [0.1, 0.2): 20168 (24.64%) [0.2, 0.3): 9409 (11.50%) [0.3, 0.4): 5732 (7.00%) [0.4, 0.5): 3826 (4.68%) [0.5, ~ ~): 9711 (11.87%)
地面 9936 points	dist < 0.1: 9 (0.09%) [0.1, 0.2): 574 (5.78%) [0.2, 0.3): 3940 (39.65%) [0.3, 0.4): 1586 (15.96%) [0.4, 0.5): 1061 (10.68%) [0.5, ~ ~): 2766 (27.84%)	dist < 0.1: 1899 (19.11%) [0.1, 0.2): 2128 (21.42%) [0.2, 0.3): 1551 (15.61%) [0.3, 0.4): 1055 (10.62%) [0.4, 0.5): 822 (8.27%) [0.5, ~ ~): 2481 (24.97%)	dist < 0.1: 1322 (13.31%) [0.1, 0.2): 2619 (26.36%) [0.2, 0.3): 1508 (15.18%) [0.3, 0.4): 1010 (10.17%) [0.4, 0.5): 805 (8.10%) [0.5, ~ ~): 2672 (26.89%)	dist < 0.1: 1774 (17.85%) [0.1, 0.2): 2207 (22.21%) [0.2, 0.3): 1496 (15.06%) [0.3, 0.4): 1057 (10.64%) [0.4, 0.5): 775 (7.80%) [0.5, ~ ~): 2627 (26.44%)
去地面点 19897 points	dist < 0.1: 1467 (7.37%) [0.1, 0.2): 4889 (24.57%) [0.2, 0.3): 5932 (29.81%) [0.3, 0.4): 3178 (15.97%) [0.4, 0.5): 1393 (7.00%) [0.5, ~ ~): 3038 (15.27%)	dist < 0.1: 3850 (19.35%) [0.1, 0.2): 7387 (37.13%) [0.2, 0.3): 3342 (16.80%) [0.3, 0.4): 1768 (8.89%) [0.4, 0.5): 1004 (5.05%) [0.5, ~ ~): 2546 (12.80%)	dist < 0.1: 5437 (27.33%) [0.1, 0.2): 5784 (29.07%) [0.2, 0.3): 2886 (14.50%) [0.3, 0.4): 1738 (8.73%) [0.4, 0.5): 1108 (5.57%) [0.5, ~ ~): 2944 (14.80%)	dist < 0.1: 5453 (27.41%) [0.1, 0.2): 5817 (29.24%) [0.2, 0.3): 2917 (14.66%) [0.3, 0.4): 1737 (8.73%) [0.4, 0.5): 1128 (5.67%) [0.5, ~ ~): 2845 (14.30%)

2. 里程计全局建图（scan-to-map）

2.1 轨迹：

点云匹配的到的位姿在局部更加精确，体现在柱面粗细、墙面厚度拟合的精度，在较大范围内z轴上漂移较大。

2.2 量化分析：

特征匹配建图的点云与静止点云的匹配精度较IMU + GPS的精度更好。
匹配后点距大于0.5m的点主要为未去除的动态物体（车辆/行人）。
高度越高，匹配精度越差。

2.2.1 TEST SCENE 1

USING ODOM	USING IMU
ODOM MAP 1,000,000 points STATIC TOTAL: 167635 points dist < 0.1: 57218 (34.13249023175351%) [0.1,0.2): 64855 (38.68822143347153%) [0.2, 0.3): 21072 (12.570167327825335%) [0.3, 0.4): 9238 (5.510782354520238%) [0.4, 0.5): 5758 (3.4348435589226596%) [0.5, ~ ~): 9494 (5.663495093506725%)	IMU MAP 1,000,000 points STATIC TOTAL: 167635 points dist < 0.1: 24193 (14.43195036835983%) [0.1, 0.2): 77132 (46.01187102931965%) [0.2, 0.3): 40690 (24.27297402093835%) [0.3, 0.4): 11747 (7.0074865034151586%) [0.4, 0.5): 4409 (2.630119008560265%) [0.5, ~ ~): 9464 (5.645599069406747%)

ODOM DISTANCE 0~1m	IMU DISTANCE MAP 0~1m

ODOM DISTANCE MAP 0～0.5m	IMU DISTANCE MAP 0~0.5m

2.2.2 TEST SCENE 2

USING ODOM	USING IMU
ODOM MAP 1,000,000 points STATIC TOTAL: 124486 points dist < 0.1: 109772 (88.1801969699404%) [0.1, 0.2): 13553 (10.887168034959755%) [0.2, 0.3): 144 (0.11567565830695821%) [0.3, 0.4): 95 (0.07631380235528494%) [0.4, 0.5): 91 (0.07310058962453608%) [0.5, ~ ~): 831 (0.6675449448130714%)	IMU MAP 1,000,000 points STATIC TOTAL: 124486 points dist < 0.1: 83008 (66.68059058849991%) [0.1, 0.2): 40045 (32.168275950709315%) [0.2, 0.3): 397 (0.3189113635268223%) [0.3, 0.4): 101 (0.08113362145140819%) [0.4, 0.5): 94 (0.07551049917259772%) [0.5, ~ ~): 841 (0.6755779766399435%)

ODOM DISTANCE 0~0.5m	IMU DISTANCE MAP 0~0.5m