Occluding Contour Detection With DUC and HDC Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional semantic segmentation systems face challenges in accurately detecting occluded objects in traffic environments due to the 'gridding issue' caused by standard dilated convolution operations and the loss of information during downsampling, which affects object contour detection and occlusion reasoning in autonomous driving systems.
Innovation Solution
The implementation of Dense Upsampling Convolution (DUC) and Hybrid Dilated Convolution (HDC) frameworks within a fully convolutional network (FCN) for pixel-level semantic segmentation, which enhances object contour detection by alleviating the gridding issue and recovering fine-detailed information, enabling accurate detection of occluded objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If standard dilated convolution operations are used to enlarge the field of convolutional kernels, then context aggregation and multi-scale object capture are improved, but a gridding issue is produced that degrades measurement precision
Solution Approach 1:
The patent applies parameter changes by modifying the convolution operation parameters - specifically using overlapping dilated convolution with adjustable dilation rates and overlap ratios. This allows the system to maintain the benefits of context aggregation while adjusting the convolution parameters to eliminate the gridding issue and improve contour detection precision.
2Productivity
If max-pooling or strided convolution is used in CNNs to downsample feature maps, then computational efficiency is improved, but information is lost that is necessary for identification of important objects
Solution Approach 1:
The patent applies preliminary action by performing object contour detection and identification before the final downsampling operation. The system detects occluded object contours at higher resolution stages, then uses this preliminary detection information to guide subsequent downsampling, ensuring that important object information is preserved even after aggressive pooling operations.
3Reliability
If bounding box merging is applied in conventional object detection frameworks to reduce false positive rate, then false positive reduction is improved, but occluded objects are neglected and undetected
Solution Approach 1:
The patent applies segmentation by dividing the object detection task into two independent components: occluding object detection and occluded object detection. Instead of merging bounding boxes that may belong to different objects, the system segments the detection process to separately identify occluding contours and occluded contours, preserving occluded object information while maintaining reliability.
4Measurement precision
If conventional semantic segmentation is applied to assign categorical labels to pixels, then pixel-level classification is improved, but object-level instance information and occlusion reasoning are lost
Solution Approach 1:
The patent applies merging by combining pixel-level semantic segmentation with object-level contour detection in a unified framework. The system simultaneously performs pixel classification and occluded object contour detection, merging the benefits of both approaches to achieve accurate pixel-level labeling while preserving object instance information and occlusion reasoning capabilities.
Data Source
AI summary
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.