Spatial aware object detection by LIDAR and camera fusion based super-resolution

1. A method for training an artificial intelligence module of an advanced driver assistance system for obtaining a depth model of a scene, the method comprising:

acquiring multiple sets of training data, each set of training data among the multiple sets of training data including a camera image of the scene taken by a camera of a vehicle, a depth image of the scene at a first resolution taken by a LIDAR sensor of the vehicle, the depth image comprising data indicating a distance of each pixel of the depth image from the LIDAR sensor, and depth measurement data of the scene at a second resolution that is higher than the first resolution taken by a plurality of other LIDAR sensors, the depth measurement data comprising data indicating a distance of each pixel of images of the plurality of other LIDAR sensors from the plurality of other LIDAR sensors;

combining the images of the plurality of other LIDAR sensors to obtain the depth measurement data of the scene from a perspective of the LIDAR sensor of the vehicle, wherein a resolution of the depth measurement data of the scene is higher than a resolution of the images of the plurality of other LIDAR sensors; and

training the artificial intelligence module of the advanced driver assistance system with the multiple sets of training data for obtaining the depth model of the scene with the second resolution if provided with the camera image and the depth image with the first resolution as inputs to the artificial intelligence model.

2. The method according to claim 1 , wherein the plurality of other LIDAR sensors of which images are combined are not of the vehicle.

3. The method according to claim 1 , further comprising preprocessing the multiple sets of training data with occlusion filtering.

4. The method according to claim 1 , wherein the depth measurement data are used as ground truth for the depth model, and wherein training the artificial intelligence module comprises calculating a first loss function regarding a supervised super-resolution loss representing a deviation of the depth model of the artificial intelligence module from the ground truth in a first field of view.

5. The method according to claim 4 , wherein training the artificial intelligence module further comprises calculating a second loss function regarding an unsupervised scene structure loss based on a structural content of the camera image of the scene in a second field of view.

6. The method according to claim 5 , wherein the first field of view is comprised within the field of view of the camera and the field of view of the LIDAR sensor, wherein the second field of view is comprised within the field of view of the camera, and wherein the depth model in the second field of view is calibrated with the depth model in the first field of view.

7. The method according to claim 1 , wherein the depth model comprises a distance of each data point of the depth model from the camera and/or from the LIDAR sensor.

8. The method according to claim 1 , wherein the depth model further comprises a reflectivity of each data point of the depth model with respect to a radiation of the LIDAR sensor.

9. The method according to claim 1 , wherein the depth model further comprises the RGB values of the camera image for each data point.

10. The method according to claim 1 , wherein the depth model of the scene comprises a resolution higher than the first resolution of the depth image of the scene of the LIDAR sensor.

11. An advanced driver assistance system, comprising:

a camera configured to capture an image of the scene;

a LIDAR sensor configured to capture a depth image of the scene at a first resolution; and

a processor configured to train an artificial intelligence module for obtaining a depth model of the scene by acquiring multiple sets of training data, each set of training data among the multiple sets of training data including a camera image of the scene taken by the camera, a depth image of the scene taken by the LIDAR sensor, the depth image comprising data indicating a distance of each pixel of the depth image from the LIDAR sensor, and depth measurement data of the scene at a second resolution that is higher than the first resolution taken by a plurality of other LIDAR sensors, the depth measurement data comprising data indicating a distance of each pixel of images of the plurality of other LIDAR sensors from the plurality of other LIDAR sensors;

combine the images of the plurality of other LIDAR sensors to obtain the depth measurement data of the scene from a perspective of the LIDAR sensor of the vehicle, wherein a resolution of the depth measurement data of the scene is higher than a resolution of the images of the plurality of other LIDAR sensors; and

wherein the artificial intelligence module is configured to output the depth model of the scene with the second resolution if provided with the camera image and the depth image with the first resolution as inputs to the artificial intelligence model.