Method of installation height estimation and method of driver assistance system algorithm update

Abstract

An installation height estimation method utilizes a computer device to perform the following steps: acquiring multiple images of the front of a vehicle; using a vehicle position and size estimation model to obtain image coordinate information, a predicted true vehicle width, and a predicted true vehicle height for a target vehicle in each image of the front of the vehicle; for each image of the front of the vehicle, based on the corresponding image coordinate information, the predicted true vehicle width, the predicted true vehicle height, a center point image coordinate, and a horizon image coordinate, using a pinhole camera model to obtain a candidate installation height from multiple preset installation heights; and using a regression analysis method to obtain a true installation height from the candidate installation heights.

Claims

1. A method for estimating the installation height of a camera module installed in a vehicle, executed using a computer device, the computer device having a pre-established vehicle position and size estimation model, the method comprising the following steps: (A) obtaining multiple images of the front of the vehicle taken by the camera module; (B) for each image of the front of the vehicle, using the vehicle position and size estimation model to obtain image coordinate information of a bounding box corresponding to a target vehicle, a predicted true vehicle width, and a predicted true vehicle height, wherein the predicted true vehicle width and the predicted true vehicle height can be converted into a set of world coordinate information corresponding to a first direction and a second direction; (C) based on the image coordinate information of the target vehicle in all the images of the front of the vehicle and the set of world coordinate information, calculating and obtaining an intrinsic parameter data matrix of the camera module; (D) for each image of the front of the vehicle, based on the data obtained in steps (B) to (C) and a pinhole camera model, obtaining an extrinsic parameter data matrix containing a pitch angle data corresponding to the camera module when taking the image of the front of the vehicle. (E) For each target vehicle in the front view, based on the data obtained in steps (B) to (D) and multiple preset installation heights, obtain multiple third-direction world coordinate values ​​corresponding to the preset installation heights and relating to the target vehicle in a third direction; (F) For each target vehicle in the front view, based on the data obtained in steps (B) to (E), reproject the target vehicle in the image coordinate system to obtain reprojected image coordinate information of the bounding box corresponding to the target vehicle and a reprojection error value; (G) For each front view, take the preset installation height corresponding to the third-direction world coordinate value with the smallest reprojection error value as a candidate installation height; and (H) Obtain a true installation height from the candidate installation heights.

2. A method for estimating the installation height of a camera module installed in a vehicle, executed using a computer device, the computer device having pre-established a vehicle position and size estimation model, the method comprising the following steps: (A) obtaining multiple images of the front of the vehicle taken by the camera module; (B) for each image of the front of the vehicle, using the vehicle position and size estimation model to obtain image coordinate information of the bounding box corresponding to a target vehicle in the image of the front of the vehicle, the predicted true width of the target vehicle, and the predicted true height of the target vehicle; (C) for each target vehicle in the image of the front of the vehicle, based on the predicted true width and predicted true height of the target vehicle, obtaining first-direction world coordinate information of the target vehicle in a first direction and second-direction world coordinate information in a second direction in a world coordinate system defined according to a pinhole camera model; (D) based on the image coordinate information of the target vehicle in all the images of the front of the vehicle, the first-direction world coordinate information, and the second-direction world coordinate information, using the pinhole camera model to obtain a lens focal length of the camera module; (E) For each front view of a vehicle, based on the lens focal length, a center point image coordinate of the image center point of the front view in the image coordinate system, and a horizon image coordinate of the position of the horizon in the front view in the image coordinate system, a pitch angle of the camera module when capturing the front view of the vehicle is obtained; (F) For each target vehicle in the front view of a vehicle, based on the pitch angle corresponding to the front view of the vehicle, the lens focal length, the horizon image coordinate of the front view of the vehicle, the image coordinate information of the target vehicle, and multiple preset installation heights, multiple third-direction world coordinate values ​​corresponding to the preset installation heights and relating to the target vehicle in the world coordinate system in a third-direction direction are obtained; (G) For each third-direction world coordinate value of the target vehicle in each front-view image, based on the first-direction world coordinate information of the target vehicle, the second-direction world coordinate information of the target vehicle, an intrinsic parameter data matrix of the camera module, and an extrinsic parameter data matrix of the front-view image, the target vehicle is reprojected in the image coordinate system to obtain reprojected image coordinate information corresponding to the bounding box of the reprojected target vehicle and a reprojection error value, wherein the reprojection error value is the error between the image coordinate information of the target vehicle and the reprojected image coordinate information; (H) For each front-view image, the preset installation height corresponding to the third-direction world coordinate value with the smallest reprojection error value is taken as a candidate installation height; and (I) A true installation height is obtained from the candidate installation heights using a regression analysis method.

3. The installation height estimation method as described in claim 1 or 2, wherein the vehicle position and size estimation model is obtained by using target detection technology and training based on a loss function relating the actual size of the vehicle to the predicted actual width and predicted actual height of the target vehicle.

4. The installation height estimation method as described in claim 2, wherein, In step (E), for each front view, its corresponding pitch angle is calculated by the following formula: where represents the pitch angle corresponding to the front view, represents the second direction image coordinate value of the center point image coordinate of the front view in the second direction, represents the second direction image coordinate value of the horizon image coordinate of the front view in the second direction, and represents the focal length of the lens.

5. The installation height estimation method as described in claim 2, wherein, In step (F), for each target vehicle in the front view, its corresponding third-party world coordinate value is calculated by the following formula: where represents the third-party world coordinate value of the target vehicle, represents the pitch angle corresponding to the front view, represents the focal length of the lens, represents any preset installation height, represents the second-direction image coordinate value of the frame line of the target vehicle's bounding box near the bottom of the target vehicle in the second direction, and represents the second-direction image coordinate value of the horizon image coordinate of the front view in the second direction.

6. The installation height estimation method as described in claim 2, wherein, In step (G), for each third-direction world coordinate value of the target vehicle in each front-view image, the corresponding reprojected image coordinate information includes the image coordinates of each vertex of the bounding box of the reprojected target vehicle corresponding to the reprojected image coordinate information, and the image coordinates of each vertex are obtained by the following formula: where represents the intrinsic data matrix, represents the extrinsic data matrix of the front-view image, represents an arbitrary scale factor, represents the first-direction world coordinate value of the vertex corresponding to the vertex contained in the first-direction world coordinate information of the target vehicle, represents the second-direction world coordinate value of the vertex corresponding to the vertex contained in the second-direction world coordinate information of the target vehicle, represents the third-direction world coordinate value of the target vehicle, represents the first-direction image coordinate value of the vertex in the first direction, and represents the second-direction image coordinate value of the vertex in the second direction.

Images