Abstract: A system and method for generating training data for fine-tuning an Object Detector (OD) are provided. The method comprises: receiving a given sequence of labelled 3D point clouds of the source domain; selecting, in the sequence bounding boxes, a quasi-stationary bounding box that is a most overlapped bounding box by other bounding boxes from the sequence of bounding boxes; generating a re-labelled 3D point cloud of the source domain by relocating, in the given labelled 3D point cloud, the respective bounding box of the training object to a location thereof corresponding to a location of the quasi-stationary bounding box in an other labelled 3D point cloud; and storing the re-labelled 3D point cloud of the source domain in an internal memory of the server.

Abstract: Methods and systems for encoding 3D data for use with 2D convolutional neural networks (CNNs) are described. A set of 3D data is encoded into a set of one or more arrays. A 2D index of the arrays is calculated by projecting 3D coordinates of the 3D point onto a 2D image plane that is defined by a set of defined virtual camera parameters. The virtual camera parameters include a camera projection matrix defining the 2D image plane. Each 3D coordinate of the point is stored in the arrays at the calculated 2D index. The set of encoded arrays is provided for input to a 2D CNN, for training or inference.

Abstract: Methods and systems for encoding 3D data for use with 2D convolutional neural networks (CNNs) are described. A set of 3D data is encoded into a set of one or more arrays. A 2D index of the arrays is calculated by projecting 3D coordinates of the 3D point onto a 2D image plane that is defined by a set of defined virtual camera parameters. The virtual camera parameters include a camera projection matrix defining the 2D image plane. Each 3D coordinate of the point is stored in the arrays at the calculated 2D index. The set of encoded arrays is provided for input to a 2D CNN, for training or inference.

Abstract: A vehicle vision system includes a plurality of cameras disposed at a vehicle and having respective fields of view exterior of the vehicle. An image processor is operable, responsive to image processing of captured image data, to synthesize captured image data to generate a reduced distortion bird's-eye view image of a region exterior and near the vehicle and encompassed by the fields of view of the cameras. A display displays the bird's-eye view image for viewing by a driver of the vehicle. The image processor processes synthesized captured image data using a lane marker sensing algorithm that utilizes recursive temporal stabilization to determine lane markings in the field of view of at least the rear camera. The system determines when the vehicle is departing from its being-traveled lane and, responsive to determination that the vehicle is departing from its being-traveled lane, generates an alert to the driver of the vehicle.

Abstract: A method for determining lane markers includes providing a camera at a vehicle so as to have a field of view exterior of the vehicle. An image processor processes a frame of image data captured by the camera to determine intensity gradient information of captured image data and to determine lane markers. Contrast values at both sides of a center region of the determined lane marker are determined via processing of the intensity gradient information. An angle of the determined lane marker relative to the direction of travel of the vehicle is determined responsive to the determined contrast values. Processing of a subsequent frame of captured image data is adjusted responsive to the determined angle of the determined lane marker relative to the direction of travel of the vehicle.

Abstract: A parking assist system of a vehicle includes a camera that, when disposed at the vehicle, has a field of view exterior of the vehicle. An image processor is operable to process image data captured by the camera to detect parking space markers indicative of a parking space and to identify empty or available parking spaces. The image processor includes a parking space detection algorithm that detects parking space markers by (i) extracting low level features from captured image data, (ii) classifying pixels as being part of a parking space line or not part of a parking space line, (iii) performing spatial line fitting to find lines in the captured images and to apply parking space geometry constraints, and (iv) detecting and selecting rectangles in the captured images.

Abstract: A method for determining lane markers includes providing a camera at a vehicle so as to have a field of view exterior of the vehicle. An image processor processes a frame of image data captured by the camera to determine intensity gradient information of captured image data and to determine lane markers. Contrast values at both sides of a center region of the determined lane marker are determined via processing of the intensity gradient information. An angle of the determined lane marker relative to the direction of travel of the vehicle is determined responsive to the determined contrast values. Processing of a subsequent frame of captured image data is adjusted responsive to the determined angle of the determined lane marker relative to the direction of travel of the vehicle.

Abstract: A vision system of a vehicle includes a camera configured to be disposed at a vehicle so as to have a field of view exterior of the vehicle. An image processor is operable to process image data captured by the camera. The image processor is operable to determine lane markers on a road on which the vehicle is traveling. The image processor processes intensity gradient information of captured image data to determine lane markers, and, responsive to processing of captured image data, the image processor is operable to detect straight or curved lane markers. The image processor is operable to adapt the processing of lane marker image data in subsequent frames of captured image data responsive to image processing of lane marker image data in previous frames of captured image data.

Abstract: A system for performing depth-based scaling of 3D content. The system comprises: 1) a content source configured to provide an input image comprising a plurality of input image objects; and 2) a processor configured to receive the input image and to receive a depth map comprising depth data associated with each of the plurality of input image objects. The processor generates an output image comprising a plurality of output image objects, wherein each of the plurality of output image objects corresponding to one of the plurality of input image objects. The processor scales a size of a first output image object relative to the size of a second output image object based on depth data associated with the first output image object and the second output image object.

Abstract: A parking assist system of a vehicle includes a camera that, when disposed at the vehicle, has a field of view exterior of the vehicle. An image processor is operable to process image data captured by the camera to detect parking space markers indicative of a parking space and to identify empty or available parking spaces. The image processor includes a parking space detection algorithm that detects parking space markers by (i) extracting low level features from captured image data, (ii) classifying pixels as being part of a parking space line or not part of a parking space line, (iii) performing spatial line fitting to find lines in the captured images and to apply parking space geometry constraints, and (iv) detecting and selecting rectangles in the captured images.

Abstract: A vehicle vision system includes a plurality of cameras disposed at a vehicle and having respective fields of view exterior of the vehicle. An image processor is operable, responsive to image processing of captured image data, to synthesize captured image data to generate a reduced distortion bird's-eye view image of a region exterior and near the vehicle and encompassed by the fields of view of the cameras. A display displays the bird's-eye view image for viewing by a driver of the vehicle. The image processor processes synthesized captured image data using a lane marker sensing algorithm that utilizes recursive temporal stabilization to determine lane markings in the field of view of at least the rear camera. The system determines when the vehicle is departing from its being-traveled lane and, responsive to determination that the vehicle is departing from its being-traveled lane, generates an alert to the driver of the vehicle.

Abstract: A vision system of a vehicle includes a camera configured to be disposed at a vehicle so as to have a field of view exterior of the vehicle. An image processor is operable to process image data captured by the camera. The image processor is operable to determine lane markers on a road on which the vehicle is traveling. The image processor processes intensity gradient information of captured image data to determine lane markers, and, responsive to processing of captured image data, the image processor is operable to detect straight or curved lane markers. The image processor is operable to adapt the processing of lane marker image data in subsequent frames of captured image data responsive to image processing of lane marker image data in previous frames of captured image data.

Abstract: A system and method for adjusting the perceived depth of stereoscopic images are provided. The system includes a disparity estimator, a disparity processor and a warping engine. The disparity estimator is configured to receive a stereoscopic image, to estimate disparities in the stereoscopic image, and to generate an estimator signal comprising the estimated disparities. The disparity processor is configured to receive the estimator signal from the disparity estimator and a depth control signal that is generated based on a user input. The disparity processor is also configured to generate a processor signal based on the estimator signal and the depth control signal. The warping engine is configured to receive the processor signal and to generate an adjusted stereoscopic image by warping the processor signal based on a model.

Abstract: A system for adjusting the perceived depth of 3D content in response to a viewer input control signal. The system comprises: 1) a content source providing an input left stereoscopic image and an input right stereoscopic image; 2) a disparity estimator to receive the input left and right stereoscopic images, detect disparities between the input left and right stereoscopic images, and generate a disparities array; and 3) processing circuitry to fill in occlusion areas associated with the disparities array and apply a scale factor to the detected disparities to thereby generate a scaled disparities array. The system further comprises a warping engine to receive the scaled disparities array and generate an output left stereoscopic image and an output right stereoscopic image. The output left and right stereoscopic images have a different perceived depth than the input left and right stereoscopic images.

Abstract: Display 105 is capable of rendering, or otherwise displaying, one or more of a standard definition (SD) image, a two-dimensional (2D), a three-dimensional image (3D) and a high definition (HD) image 110.

Abstract: An image processing system includes an image reconstruction unit. The image reconstruction unit is configured to receive an image at a first resolution, apply the image to a look-up table and output a version of the image at a second resolution. The second resolution includes a higher resolution than the first resolution. In addition, the look-up table is generated inputting a plurality of training images; classifying, into a number of classes, a plurality of images patches corresponding to each of the plurality of training images; re-classifying the number of classes into a final class; and synthesizing filters corresponding to each of the class into a final filter value.

Abstract: A system for performing depth-based scaling of 3D content. The system comprises: 1) a content source configured to provide an input image comprising a plurality of input image objects; and 2) a processor configured to receive the input image and to receive a depth map comprising depth data associated with each of the plurality of input image objects. The processor generates an output image comprising a plurality of output image objects, wherein each of the plurality of output image objects corresponding to one of the plurality of input image objects. The processor scales a size of a first output image object relative to the size of a second output image object based on depth data associated with the first output image object and the second output image object.

Abstract: A system and method for adjusting the perceived depth of stereoscopic images are provided. The system includes a disparity estimator, a disparity processor and a warping engine. The disparity estimator is configured to receive a stereoscopic image, to estimate disparities in the stereoscopic image, and to generate an estimator signal comprising the estimated disparities. The disparity processor is configured to receive the estimator signal from the disparity estimator and a depth control signal that is generated based on a user input. The disparity processor is also configured to generate a processor signal based on the estimator signal and the depth control signal. The warping engine is configured to receive the processor signal and to generate an adjusted stereoscopic image by warping the processor signal based on a model.

Abstract: Display 105 is capable of rendering, or otherwise displaying, one or more of a standard definition (SD) image, a two-dimensional (2D), a three-dimensional image (3D) and a high definition (HD) image 110

Abstract: An image processing system includes an image reconstruction unit. The image reconstruction unit is configured to receive an image at a first resolution, apply the image to a look-up table and output a version of the image at a second resolution. The second resolution includes a higher resolution than the first resolution. In addition, the look-up table is generated inputting a plurality of training images; classifying, into a number of classes, a plurality of images patches corresponding to each of the plurality of training images; re-classifying the number of classes into a final class; and synthesizing filters corresponding to each of the class into a final filter value.