search

mbrossar / ai-imu-dr

AI-IMU Dead-Reckoning
MIT License
788 stars 222 forks source link

How to draw the trajectory from IMU data #43

Closed Iqun1314 closed 4 years ago

Iqun1314 commented 4 years ago

Hello ， in this paper AI-IMU Dead-Reckoning, the Fig. 1 in this paper. I want to know how to draw the pure IMU integration (cyan) . Thank you !

Iqun1314 commented 4 years ago

Hi, I have downloaded the .tex format of the paper . 1- I notice the picture named result on seq. xx of the paper. The trajectory of IMU(cyan) , proposed(green) and so on are all draw by LaTeX tool—tikzpicture. And the data is saved as .txt file. I want to know how to get(calculate) the .txt data, could you tell me the method in detail? 2- I download reformated pickle format of the 00-11 KITTI IMU raw data of yours. The data as follows: extract Could you tell me these variables represented which value in the paper, thank you so much! 3- And I also want to know the meaning of these variables: the file _deltap.p: list_rpe_validation, list_rpe, the file _normalizefactors.p: num_data, normalize_factors, u_std, u_loc . Thank you! I'm looking forward your reply.

mbrossar commented 4 years ago

The .txt file gather data from input and estimates, and it contains no post-processing.

The variables in pickle data are: Rot_c_i: rotation between car and IMU t_c_i: translation between car and IMU Rot: rotation matrix of IMU p: position of IMU b_omega: gyro bias b_acc: accelerometer bias t: time v: velocity of IMU measurements_covs : the diagonal of cov(y) given by the neural network.

list_rpe and list_rpe_validation contain a list of relation pose error

In normalize_factor, num_date are the number of points, i.e. IMU measurement, of the dataset. U_std and u_loc in normalize_factors are IMU input standard deviation and mean to learn with normalized data.

Iqun1314 commented 4 years ago

Thank you for such a detailed answer.

Iqun1314 commented 4 years ago

Hi, I have a question of your code:

delta_p = Rot[idxs_0].transpose(-1, -2).matmul( ((p[idxs_end] - p[idxs_0]).float()).unsqueeze(-1)).squeeze()

I want to know the formula of delta_p. Could you show me. And I also want to know the meaning of delta_p.

Thank you so much !

mbrossar commented 4 years ago

Hi,

delta_p is the relative position shift between instants idxs_end and idxs_0. The code is quite strange because it compute a batch of delta_p.

The reason of why having a metric based on delta_p is well explained in the "A Tutorial on Quantitative Trajectory Evaluationfor Visual(-Inertial) Odometry" (delta_p is computed in equation (23), paper link).

All the best,

Martin

Iqun1314 commented 4 years ago

Thank you so much ! 

兰德智 landz1995@foxmail.com

哈尔滨工业大学（深圳）

 

------------------ 原始邮件 ------------------ 发件人: "mbrossar/ai-imu-dr" <notifications@github.com>; 发送时间: 2020年7月26日(星期天) 晚上9:35 收件人: "mbrossar/ai-imu-dr"<ai-imu-dr@noreply.github.com>; 抄送: "安生"<610181747@qq.com>;"Author"<author@noreply.github.com>; 主题: Re: [mbrossar/ai-imu-dr] How to draw the trajectory from IMU data (#43)

Hi,

delta_p is the relative position shift between instants idxs_end and idxs_0. The code is quite strange because it compute a batch of delta_p.

The reason of why having a metric based on delta_p is well explained in the "A Tutorial on Quantitative Trajectory Evaluationfor Visual(-Inertial) Odometry" (delta_p is computed in equation (23), paper link).

All the best,

Martin

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.