Dus Operating Inc.

New Eagle Corporation

Other

Iyad Faisal Ghazi Mansour

Akhil Umat

Heinz Bodo Seifert

Kalash Jain

Indraneel Krishna Page

A method of predicting lane line types utilizing a heterogeneous convolutional neural network (HCNN) includes capturing an input image with one or more optical sensors disposed on a host member, passing the input image through the HCNN, the HCNN having at least three distinct sub-networks, the three distinct sub-networks: predicting object locations in the input image with a first sub-network; predicting lane line locations in the input image with a second sub-network; and predicting lane line types for each predicted lane line in the input image with a third sub-network.

Use of HCNN to predict lane lines types

A method of localizing a host member through sensor fusion includes capturing an input image with one or more optical sensors disposed on the host member and determining a location of the host member through a global positioning system (GPS) input. The method tracks movement of the host member through an inertial measurement unit (IMU) input, generates coordinates for the host member from the GPS input and the IMU input. The method compares the input image and a high definition (HD) map input to verify distances from the host member to predetermined objects within the input image and within the HD map input. The method continuously localizes the host member by fusing the GPS input, the IMU input, the input image, and the HD map input.

Localization of autonomous vehicles using camera, GPS, and IMU

A method of training a heterogeneous convolutional neural network (HCNN) system includes identifying batch sizes for a first task and a second task, defining images for a first batch, a second batch, and a batch x for the first task, defining images for a first batch, a second batch, and a batch y for the second task, training the HCNN using the first batch for the first task, training the HCNN using the first batch for the second task, training the HCNN using the second batch for the first task, training the HCNN using the second batch for the second task. The sequential training continues for each of the batches and each of the tasks until the end of an epoch. When the epoch is complete, the images for each batch and each task are reshuffled.

Sequential training method for heterogeneous convolutional neural network

A system and method for facial and gesture recognition, and more particularly a system and method for facial and gesture recognition using a heterogeneous convolutional neural network (CNN).

Vehicle control with facial and gesture recognition using a convolutional neural network

A heterogeneous convolutional neural network (HCNN) system includes a visual reception system generating an input image. A feature extraction layer (FEL) portion of convolutional neural networks includes multiple convolution, pooling and activation layers stacked together. The FEL includes multiple stacked layers, a first set of layers learning to represent data in a simple form including horizontal and vertical lines and blobs of colors. Following layers capture more complex shapes such as circles, rectangles, and triangles. Subsequent layers pick up complex feature combinations to form a representation including wheels, faces and grids. The FEL portion outputs data to each of: a first sub-network which performs a first task of object detection, classification, and localization for classes of objects in the input image to create a detected object table; and a second sub-network which performs a second task of defining a pixel level segmentation to create a segmentation data set.

Heterogeneous convolutional neural network for multi-problem solving

A system and method for predicting object detection and lane detection for a motor vehicle includes a convolution neural network (CNN) that receives an input image and a lane line module. The CNN includes a set of convolution and pooling layers (CPL's) trained to detect objects and lane markings from the input image, the objects categorized into object classes and the lane markings categorized into lane marking classes to generate a number of feature maps, a fully connected layer that receives the feature maps, the fully connected layer generating multiple object bounding box predictions for each of the object classes and multiple lane bounding box predictions for each of the lane marking classes from the feature maps, and a non-maximum suppression layer generating a final object bounding box prediction for each of the object classes and generating multiple final lane bounding box predictions for each of the lane marking classes.

Growing community of inventors

Auburn Hills, MI, United States of America

Iyad Faisal Ghazi Mansour