Object Contour Refinement Using Weighted Sensor Detections
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining object contours from range sensor data, such as radar or Lidar, are computationally expensive and do not accurately represent the real object boundary due to noise and clutter, often overestimating outlier points and resulting in fuzzy contours.
Innovation Solution
A method that determines an initial contour based on sensor detection subsets, associates a surrounding set of detections with each segment, and refines segments using weights based on relative positions to improve accuracy, incorporating the spread of detections and excluding outliers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If known methods (convex hull, concave hull, recasting alpha shapes, or Delaunay triangulation) are used to determine contour from point cloud, then a contour can be localized, but the computational cost is high and the accuracy is reduced due to noise and clutter causing fuzzy contours
Solution Approach 1:
The method segments the point cloud data by associating each contour segment with a specific subset of sensor detections (points). This segmentation allows the algorithm to process only relevant points for each segment rather than the entire point cloud, reducing computational complexity while maintaining contour localization accuracy.
Solution Approach 2:
The patent applies local quality by assigning different weights to different sensor detections based on their relative position to each segment. Points closer to a segment receive higher weights, allowing the algorithm to focus computational effort on locally relevant data, thereby improving measurement precision without uniformly processing all points at high cost.
2Measurement precision
If known methods connect closest points to localize contour, then a contour can be determined, but outlier points are overestimated and edges are represented by fuzzy contours
Solution Approach 1:
The method changes the parameter of point evaluation by introducing a weighting scheme based on relative position. Instead of treating all points equally or simply connecting closest points, the algorithm transforms the raw point data into weighted contributions, where the weight parameter reflects the point's relevance to each segment. This allows outlier points to be downweighted or excluded, improving both boundary representation accuracy and outlier rejection capability.
Solution Approach 2:
The patent implements feedback by using the spatial distribution information of points to adjust the weighting and refinement of contour segments. The algorithm continuously refines segment positions based on the weighted distribution of associated points, creating a feedback loop that improves contour accuracy while naturally suppressing outlier influence through the weighting mechanism.
3Measurement precision
If sensor detections with spread and noise are processed by known methods, then a contour can be generated, but the real object boundary is not represented accurately
Solution Approach 1:
The patent applies local quality by treating different regions of the point cloud differently through segment-specific weighting. Each contour segment has its own associated points with weights determined by their relative position to that segment. This localized processing approach improves object boundary accuracy by focusing computational resources on locally relevant data rather than applying uniform processing to the entire noisy point cloud.
Solution Approach 2:
The method segments both the contour and the point cloud data, associating specific subsets of sensor detections with specific contour segments. This segmentation strategy reduces processing complexity by dividing the problem into smaller, independent sub-problems, where each segment can be refined using only its associated points rather than processing the entire point cloud.
Data Source
Figure 1~1C
Figure 2~2A
Figure 3~3C
AI summary
A method is provided for determining a contour of an object with respect to a sensor being configured to provide sensor detections, each sensor detection including a respective position at the object. An initial contour including a plurality of segments is determined, each segment being related to a respective initial subset of the sensor detections. A respective surrounding set of the sensor detections is associated with each segment of the initial contour. A respective weight is provided to each of the sensor detections, each weight depending from a relative position of the sensor detection with respect to the associated segment. Each segment of the initial contour is refined by using the weights of the sensor detections of the respective surrounding set associated with the respective segment in order to determine a final contour including the refined segments for the object.