Abstract: A system and method for aerial video traffic analysis are disclosed. A particular embodiment is configured to: receive a captured video image sequence from an unmanned aerial vehicle (UAV); clip the video image sequence by removing unnecessary images; stabilize the video image sequence by choosing a reference image and adjusting other images to the reference image; extract a background image of the video image sequence for vehicle segmentation; perform vehicle segmentation to identify vehicles in the video image sequence on a pixel by pixel basis; determine a centroid, heading, and rectangular shape of each identified vehicle; perform vehicle tracking to detect a same identified vehicle in multiple image frames of the video image sequence; and produce output and visualization of the video image sequence including a combination of the background image and the images of each identified vehicle.

Abstract: A system and method for multitask processing for autonomous vehicle computation and control are disclosed.

Abstract: A system and method for instance-level lane detection for autonomous vehicle control are disclosed.

Abstract: A system method for detecting taillight signals of a vehicle using a convolutional neural network is disclosed.

Abstract: A system and method for using triplet loss for proposal free instance-wise semantic segmentation for lane detection are disclosed. A particular embodiment includes: receiving image data from an image generating device mounted on an autonomous vehicle; performing a semantic segmentation operation or other object detection on the received image data to identify and label objects in the image data with object category labels on a per-pixel basis and producing corresponding semantic segmentation prediction data; performing a triplet loss calculation operation using the semantic segmentation prediction data to identify different instances of objects with similar object category labels found in the image data; and determining an appropriate vehicle control action for the autonomous vehicle based on the different instances of objects identified in the image data.

Abstract: A system and method for vehicle occlusion detection is disclosed.

Abstract: A system method for occluding contour detection using a fully convolutional neural network is disclosed. A particular embodiment includes: receiving an input image; producing a feature map from the input image by semantic segmentation; learning an array of upscaling filters to upscale the feature map into a final dense feature map of a desired size; applying the array of upscaling filters to the feature map to produce contour information of objects and object instances detected in the input image; and applying the contour information onto the input image.

Abstract: A system and method for actively selecting and labeling images for semantic segmentation are disclosed. A particular embodiment includes: receiving image data from an image generating device; performing semantic segmentation or other object detection on the received image data to identify and label objects in the image data and produce semantic label image data; determining the quality of the semantic label image data based on prediction probabilities associated with regions or portions of the image; and identifying a region or portion of the image for manual labeling if an associated prediction probability is below a pre-determined threshold.

Abstract: A method of generating a ground truth dataset for motion planning of a vehicle is disclosed. The method includes: obtaining undistorted LiDAR scans; identifying, for a pair of undistorted LiDAR scans, points belonging to a static object in an environment; aligning the close points based on pose estimates; and transforming a reference scan that is close in time to a target undistorted LiDAR scan so as to align the reference scan with the target undistorted LiDAR scan.

Abstract: A method of generating a ground truth dataset for motion planning of a vehicle is disclosed. The method includes: corresponding, for each pair of images, a first image of the pair to a LiDAR static-scene point cloud; and computing a camera pose associated with the pair of images in the coordinate of the point cloud.

Abstract: A system for generating a ground truth dataset for motion planning of a vehicle is disclosed. The system includes an internet server that further includes an I/O port, configured to transmit and receive electrical signals to and from a client device; a memory; one or more processing units; and one or more programs stored in the memory and configured for execution by the one or more processing units, the one or more programs including instructions for: an identifying module configured to identify, for a pair of undistorted LiDAR scans, points belonging to a static object in an environment; an aligning module configured to align close points based on pose estimates; and a transforming module configured to transform a reference scan that is close in time to a target undistorted LiDAR scan so as to align the reference scan with the target undistorted LiDAR scan.

Abstract: A system for generating a ground truth dataset for motion planning of a vehicle is disclosed. The system includes an internet server that further includes an I/O port, configured to transmit and receive electrical signals to and from a client device; a memory; one or more processing units; and one or more programs stored in the memory and configured for execution by the one or more processing units, the one or more programs including instructions for: a corresponding module configured to correspond, for each pair of images, a first image of the pair to a LiDAR static-scene point cloud; and a computing module configured to compute a camera pose associated with the pair of images in the coordinate of the point cloud.

Abstract: A system and method for semantic segmentation using hybrid dilated convolution (HDC) are disclosed. A particular embodiment includes: receiving an input image; producing a feature map from the input image; performing a convolution operation on the feature map and producing multiple convolution layers; grouping the multiple convolution layers into a plurality of groups; applying different dilation rates for different convolution layers in a single group of the plurality of groups; and applying a same dilation rate setting across all groups of the plurality of groups.

Abstract: A system and method for image localization based on semantic segmentation are disclosed. A particular embodiment includes: receiving image data from an image generating device mounted on an autonomous vehicle; performing semantic segmentation or other object detection on the received image data to identify and label objects in the image data and produce semantic label image data; identifying extraneous objects in the semantic label image data; removing the extraneous objects from the semantic label image data; comparing the semantic label image data to a baseline semantic label map; and determining a vehicle location of the autonomous vehicle based on information in a matching baseline semantic label map.

Abstract: A system and method for transitioning between an autonomous and manual driving mode based on detection of a driver's capacity to control a vehicle are disclosed. A particular embodiment includes: receiving sensor data related to a vehicle driver's capacity to take manual control of an autonomous vehicle; determining, based on the sensor data, if the driver has the capacity to take manual control of the autonomous vehicle, the determining including prompting the driver to perform an action or provide an input; and outputting a vehicle control transition signal to a vehicle subsystem to cause the vehicle subsystem to take action based on the driver's capacity to take manual control of the autonomous vehicle.

Abstract: A system and method for semantic segmentation using hybrid dilated convolution (HDC) are disclosed. A particular embodiment includes: receiving an input image; producing a feature map from the input image; performing a convolution operation on the feature map and producing multiple convolution layers; grouping the multiple convolution layers into a plurality of groups; applying different dilation rates for different convolution layers in a single group of the plurality of groups; and applying a same dilation rate setting across all groups of the plurality of groups.

Abstract: A system method for occluding contour detection using a fully convolutional neural network is disclosed. A particular embodiment includes: receiving an input image; producing a feature map from the input image by semantic segmentation; applying a Dense Upsampling Convolution (DUC) operation on the feature map to produce contour information of objects and object instances detected in the input image; and applying the contour information onto the input image.

Abstract: A system and method for vehicle wheel detection is disclosed. A particular embodiment can be configured to: receive training image data from a training image data collection system; obtain ground truth data corresponding to the training image data; perform a training phase to train one or more classifiers for processing images of the training image data to detect vehicle wheel objects in the images of the training image data; receive operational image data from an image data collection system associated with an autonomous vehicle; and perform an operational phase including applying the trained one or more classifiers to extract vehicle wheel objects from the operational image data and produce vehicle wheel object data.

Abstract: A system method for occluding contour detection using a fully convolutional neural network is disclosed. A particular embodiment includes: receiving an input image; producing a feature map from the input image by semantic segmentation; applying a Dense Upsampling Convolution (DUC) operation on the feature map to produce contour information of objects and object instances detected in the input image; and applying the contour information onto the input image.

Abstract: A system and method for semantic segmentation using dense upsampling convolution (DUC) are disclosed. A particular embodiment includes: receiving an input image; producing a feature map from the input image; performing a convolution operation on the feature map and reshape the feature map to produce a label map; dividing the label map into equal subparts, which have the same height and width as the feature map; stacking the subparts of the label map to produce a whole label map; and applying a convolution operation directly between the feature map and the whole label map without inserting extra values in deconvolutional layers to produce a semantic label map.