System and method for occluding contour detection

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.

What is claimed is: 
     
         1 . A system comprising:
 a data processor;   a trained computational model trained with ground truth data comprising object data, object occluding contour data, and training labels, the trained computational model being further trained to learn an array of upscaling filters to upscale a downsized feature map into a final dense feature map of a given size; and
 an occluding object contour detection processing module, executable by the data processor, the occluding object contour detection processing module being configured to at least:
 generate, based on the final dense feature map, contour information of object instances that appear in an input image of the given size. 
 
   
     
     
         2 . The system of  claim 1  wherein the downsized feature map is produced from the input image by semantic segmentation. 
     
     
         3 . The system of  claim 1  wherein the downsized feature map is produced from the input image by semantic segmentation, wherein the semantic segmentation is performed by a deep convolutional neural network trained on a dataset configured for a traffic environment. 
     
     
         4 . The system of  claim 1  being configured to operate within a fully convolutional network. 
     
     
         5 . The system of  claim 1  wherein the contour information is generated without the use of bounding boxes. 
     
     
         6 . The system of  claim 1  further including an autonomous vehicle control subsystem, which uses the contour information to control a vehicle without a driver. 
     
     
         7 . The system of  claim 1  being further configured to use dense upsampling convolution (DUC) to generate pixel-level predictions of objects detected in the input image. 
     
     
         8 . A method comprising:
 using a trained computational model, trained with ground truth data comprising object data, object occluding contour data, and training labels, to learn an array of upscaling filters to upscale a downsized feature map into a final dense feature map of a given size; and   using an occluding object contour detection processing module, executable by a data processor, to generate, based on the final dense feature map, contour information of object instances that appear in an input image of the given size.   
     
     
         9 . The method of  claim 8  including applying different dilation rates to the downsized feature map. 
     
     
         10 . The method of  claim 8  including applying convolutional operations on the downsized feature map in multiple convolutional layers. 
     
     
         11 . The method of  claim 8  including using dense upsampling convolution with semantic segmentation. 
     
     
         12 . The method of  claim 8  including applying a range of different dilation rates using hybrid dilation convolution with semantic segmentation. 
     
     
         13 . The method of  claim 8  including using the contour information with an autonomous vehicle motion planner to control a vehicle without a driver. 
     
     
         14 . The method of  claim 8  including using dense upsampling convolution to generate pixel-level predictions of objects detected in the input image. 
     
     
         15 . A non-transitory machine-useable storage medium embodying instructions which, when executed by at least one processor, cause the at least one processor to at least:
 use a trained computational model, trained with ground truth data comprising object data, object occluding contour data, and training labels, to learn an array of upscaling filters to upscale a downsized feature map into a final dense feature map of a given size; and   use an occluding object contour detection processing module, executable by a data processor, to generate, based on the final dense feature map, contour information of object instances that appear in an input image of the given size.   
     
     
         16 . The non-transitory machine-useable storage medium of  claim 15  wherein the instructions are further configured to apply a range of different dilation rates as part of a convolution operation. 
     
     
         17 . The non-transitory machine-useable storage medium of  claim 15  wherein the instructions are further configured to use a deep convolutional neural network trained on a cityscape dataset to perform semantic segmentation on the input image. 
     
     
         18 . The non-transitory machine-useable storage medium of  claim 15  wherein the instructions are further configured to use conditional random fields. 
     
     
         19 . The non-transitory machine-useable storage medium of  claim 15  wherein the instructions are further configured to generate the contour information in addition to a use of bounding boxes. 
     
     
         20 . The non-transitory machine-useable storage medium of  claim 15  wherein the instructions are further configured to use the contour information with an autonomous vehicle motion planner to plan a route for an autonomous vehicle.