After reducing the size of the large image to the image size (752 * 480) of the Euroc dataset, the output result became worse

[SerpentWindy]

My image size is 1164 * 874, and the image includes the front hood of the car. During the one kilometer journey, VINS did not experience a system reboot, but the trajectory distance and angle were inaccurate. Then I resized and cropped the images, and ran VINS again using the modified images as the dataset. However, there was a "big translation" system reboot error during the run. I want to know why the result is worse after resizing and cutting ? The reason why I resized and collapsed the image is as follows:

1. I think the front hatch will affect the effectiveness of feature tracking and visual modeling, so I tried to crop the area of the front hatch.
2. I think the feature detection parameters I use are all parameters corresponding to the Euroc dataset, and my image size is too large compared to (752,480). These feature detection parameters are not suitable for my image, so I want to resize the image with a size of 1164 * 874 to (752,480).

In summary, I resized and collapsed the 1164 874 image to obtain a 752 480 size image, with the front hatch removed. Because I changed the image size, I needed to modify the camera intrinsic matrix. Without changing other parameters, I encountered a "big translation" system reboot error while running VINS,

My modification process is as follows:

1.   I resized the image with dimensions of 1164 * 874 to (752,600),
2.   crop the (752,600) size image to (752,480) size,
3.   a new bag was created using  (752,480) images and IMU data,
4.   modify the camera's intrinsic matrix based on my resizing and cropping.

My original fx=910.0, fy=910.0, cx=582.0, cy=437.0, After resize, fx=910.0752/1164=587.90378, fy=910.0600/874=624.71395881, cx=582.0752/1164=376.0, cy=437.0600/874=300, After crop, fx=587.90378, fy=624.71395881, cx=376.0, cy=300-(600-480)/2=240.0

[SerpentWindy]

This is my original image

[SerpentWindy]

This is my modified image

[SerpentWindy]

This is my original parameter table： 

camera calibration

model_type: PINHOLE camera_name: camera  image_width: 1164 image_height: 874 distortion_parameters: k1: 0.1322964747 k2: -0.2047703774 p1: 0.0000790454 p2: 0.00001595080 projection_parameters: fx: 9.1e+02 fy: 9.1e+02 cx: 5.82e+02 cy: 4.37e+02 

Extrinsic parameter between IMU and Camera.

estimate_extrinsic: 1 #0 Have an accurate extrinsic parameters. We will trust the following imu^R_cam, imu^T_cam, don't change it. 1 Have an initial guess about extrinsic parameters. We will optimize around your initial guess. 2 Don't know anything about extrinsic parameters. You don't need to give R,T. We will try to calibrate it. Do some rotation movement at beginning.

If you choose 0 or 1, you should write down the following matrix.

Rotation from camera frame to imu frame, imu^R_cam

extrinsicRotation: !!opencv-matrix rows: 3 cols: 3 dt: d data: [0.0017453, -0.0017423, 0.9999970, 0.9999970, 0.0017484, -0.0017423, -0.0017453, 0.9999970, 0.0017453]

Translation from camera frame to imu frame, imu^T_cam

extrinsicTranslation: !!opencv-matrix rows: 3 cols: 1 dt: d data: [0.01,0.01,0.01] 

feature traker paprameters

max_cnt: 150 # max feature number in feature tracking min_dist: 30 # min distance between two features freq: 10 # frequence (Hz) of publish tracking result. At least 10Hz for good estimation. If set 0, the frequence will be same as raw image F_threshold: 1.0 # ransac threshold (pixel) show_track: 1 # publish tracking image as topic equalize: 1 # if image is too dark or light, trun on equalize to find enough features fisheye: 0 # if using fisheye, trun on it. A circle mask will be loaded to remove edge noisy points 

optimization parameters

max_solver_time: 0.04 # max solver itration time (ms), to guarantee real time max_num_iterations: 8 # max solver itrations, to guarantee real time keyframe_parallax: 10.0 # keyframe selection threshold (pixel) 

imu parameters The more accurate parameters you provide, the better performance

acc_n: 0.08 # accelerometer measurement noise standard deviation. #0.2 0.04 gyr_n: 0.004 # gyroscope measurement noise standard deviation. #0.05 0.004 acc_w: 0.00004 # accelerometer bias random work noise standard deviation. #0.02 gyr_w: 2.0e-6 # gyroscope bias random work noise standard deviation. #4.0e-5 g_norm: 9.81007 # gravity magnitude 

loop closure parameters

loop_closure: 1 # start loop closure load_previous_pose_graph: 0 # load and reuse previous pose graph; load from 'pose_graph_save_path' fast_relocalization: 0 # useful in real-time and large project pose_graph_save_path: "/home/nvidia/vio_output/pose_graph/" # save and load path 

unsynchronization parameters

estimate_td: 0 # online estimate time offset between camera and imu td: 0.0 # initial value of time offset. unit: s. readed image clock + td = real image clock (IMU clock) 

rolling shutter parameters

rolling_shutter: 1 # 0: global shutter camera, 1: rolling shutter camera rolling_shutter_tr: 0.029 # unit: s. rolling shutter read out time per frame (from data sheet). 

visualization parameters

save_image: 1 # save image in pose graph for visualization prupose; you can close this function by setting 0 visualize_imu_forward: 0 # output imu forward propogation to achieve low latency and high frequence results visualize_camera_size: 0.4 # size of camera marker in RVIZ  

[SerpentWindy]

This is my modified parameter table： 

camera calibration

model_type: PINHOLE camera_name: camera  image_width: 752 image_height: 480

distortion_parameters:

k1: 0.1322964747 k2: -0.2047703774 p1: 0.0000790454 p2: 0.00001595080

projection_parameters:

fx: 587.90378 fy: 624.71395881 cx: 376.0 cy: 240.0 

Extrinsic parameter between IMU and Camera.

estimate_extrinsic: 1 #0 Have an accurate extrinsic parameters. We will trust the following imu^R_cam, imu^T_cam, don't change it. 1 Have an initial guess about extrinsic parameters. We will optimize around your initial guess. 2 Don't know anything about extrinsic parameters. You don't need to give R,T. We will try to calibrate it. Do some rotation movement at beginning.

If you choose 0 or 1, you should write down the following matrix.

Rotation from camera frame to imu frame, imu^R_cam

extrinsicRotation: !!opencv-matrix rows: 3 cols: 3 dt: d data: [0.0017453, -0.0017423, 0.9999970, 0.9999970, 0.0017484, -0.0017423, -0.0017453, 0.9999970, 0.0017453]

Translation from camera frame to imu frame, imu^T_cam

extrinsicTranslation: !!opencv-matrix rows: 3 cols: 1 dt: d data: [0.01,0.01,0.01] 

feature traker paprameters

max_cnt: 150 # max feature number in feature tracking min_dist: 30 # min distance between two features freq: 10 # frequence (Hz) of publish tracking result. At least 10Hz for good estimation. If set 0, the frequence will be same as raw image F_threshold: 1.0 # ransac threshold (pixel) show_track: 1 # publish tracking image as topic equalize: 1 # if image is too dark or light, trun on equalize to find enough features fisheye: 0 # if using fisheye, trun on it. A circle mask will be loaded to remove edge noisy points 

optimization parameters

max_solver_time: 0.04 # max solver itration time (ms), to guarantee real time max_num_iterations: 8 # max solver itrations, to guarantee real time keyframe_parallax: 10.0 # keyframe selection threshold (pixel) 

imu parameters The more accurate parameters you provide, the better performance

acc_n: 0.08 # accelerometer measurement noise standard deviation. #0.2 0.04 gyr_n: 0.004 # gyroscope measurement noise standard deviation. #0.05 0.004 acc_w: 0.00004 # accelerometer bias random work noise standard deviation. #0.02 gyr_w: 2.0e-6 # gyroscope bias random work noise standard deviation. #4.0e-5 g_norm: 9.81007 # gravity magnitude 

loop closure parameters

loop_closure: 1 # start loop closure load_previous_pose_graph: 0 # load and reuse previous pose graph; load from 'pose_graph_save_path' fast_relocalization: 0 # useful in real-time and large project pose_graph_save_path: "/home/nvidia/vio_output/pose_graph/" # save and load path 

unsynchronization parameters

estimate_td: 0 # online estimate time offset between camera and imu td: 0.0 # initial value of time offset. unit: s. readed image clock + td = real image clock (IMU clock) 

rolling shutter parameters

rolling_shutter: 1 # 0: global shutter camera, 1: rolling shutter camera rolling_shutter_tr: 0.029 # unit: s. rolling shutter read out time per frame (from data sheet). 

visualization parameters

save_image: 1 # save image in pose graph for visualization prupose; you can close this function by setting 0 visualize_imu_forward: 0 # output imu forward propogation to achieve low latency and high frequence results visualize_camera_size: 0.4 # size of camera marker in RVIZ