Problem about using multi-camera

[MrlynnBei]

Hi, thank you for sharing the great work!

Our vehicle is equipped with six cameras, mounted on the front, back, left, and right, each with a different height. I have tested cameras with similar heights and achieved good results. However, I found that when I use all the cameras, some of them have significantly different heights, resulting in poor performance.

The code below only uses one camera height to estimate flat plane. I would like to know how I should modify the code to adapt to these multiple cameras？Any reply would be appreciated.

` # 6. estimate flat plane self.file_check() self.label_valid_check() ref_camera2world_all = np.array(self.ref_camera2world_all) transform_normal2origin = robust_estimate_flatplane(np.array(ref_camera2world_all)[:, :3, 3]).astype(np.float32) transform_normal2origin[0, 3] = -self.choose_pt[0] transform_normal2origin[1, 3] = -self.choose_pt[1] transform_normal2origin[2, 3] += self.camera_height self.ref_camera2world_all = transform_normal2origin[None] @ self.ref_camera2world_all

[DRosemei]

@MrlynnBei Hi, you can refer https://github.com/DRosemei/RoMe/issues/42 for help. The estimated flat plane is not always good.

[MrlynnBei]

@MrlynnBei Hi, you can refer #42 for help. The estimated flat plane is not always good.

Okk. Thanks! And I have another question: Can the world coordinate system of the camera pose be an ENU system? or Can the camera pose be expressed in ENU coordinates?（tf_enu_cami）？

I saw the code that transform the tf_cam0_cami to tf_chassis_cami：(Am I understanding this correctly?) camera_poses = self.camera2chassis @ camera_poses # refrece to chassis

Any reply would be appreciated.

[DRosemei]

@MrlynnBei We use local corrdinate systems and save a tranform from local to global coordinate. It would be better for understanding and reconstruction.