Lidar Obstacle Detection via Feature Map Weighting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current lidar-based obstacle detection methods in autonomous driving face challenges such as high resource consumption, missed detection, and false detection due to the complexity of point cloud data, with existing methods failing to effectively extract obstacle features.
Innovation Solution
A method that converts lidar point cloud data into an obstacle feature map using a neural network structure, dividing the feature extraction process into encoding and decoding stages with attention mechanisms and supervised learning, to enhance the identification and detection of obstacles by increasing weight values and incorporating monitoring data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If the obstacle detection method based on point cloud segmentation is used, then the detection range is large, but the detection precision is reduced and specific obstacle detection results cannot be obtained
Solution Approach 1:
The patent segments the point cloud data processing into distinct stages: initial point cloud segmentation for broad coverage, followed by lattice distribution for structured organization, and finally focused feature extraction for precise obstacle identification. This multi-stage segmentation approach maintains large detection range while progressively improving detection precision at each level.
Solution Approach 2:
The patent transforms point cloud data from 3D spatial coordinates into 2D lattice distributions and further into feature maps, adding semantic dimensionality to the spatial data. This dimensional transformation enables the system to maintain comprehensive spatial coverage while extracting precise obstacle features through the added semantic layer.
2Measurement precision
If the obstacle detection method based on points is used, then the detection precision is high, but the resource consumption is excessive
Solution Approach 1:
The patent extracts only the essential features from the complete point cloud data through a hierarchical processing approach. Instead of analyzing all points in detail, it first identifies regions of interest through lattice distribution, then extracts features only from those specific regions, significantly reducing computational resources while maintaining high detection precision for actual obstacles.
Solution Approach 2:
The patent applies partial processing to the point cloud data by focusing computational effort only on relevant regions identified through lattice distribution and weight value adjustment. Rather than performing exhaustive analysis on all points, it processes only the necessary portions at each stage, reducing overall resource consumption while maintaining sufficient detection precision.
3Use of energy by moving object
If the obstacle detection method based on lattice distribution is used, then the resource consumption is reduced, but missed detection of obstacles occurs
Solution Approach 1:
The patent implements feedback mechanisms through weight value adjustment based on point cloud distribution characteristics. The system continuously monitors detection results and adjusts the weights of different lattice regions dynamically, reinforcing areas with higher obstacle probability. This feedback loop reduces missed detections while maintaining efficient resource consumption through the lattice structure.
Solution Approach 2:
The patent changes key parameters during the detection process, including weight values assigned to different lattice regions, threshold values for obstacle identification, and resolution levels at different processing stages. These parameter adjustments allow the system to optimize between resource consumption and detection reliability dynamically, reducing missed detections without sacrificing efficiency.
4Measurement precision
If the feature extraction process processes all point cloud data in detail, then the detection precision is high, but the processing time is excessive
Solution Approach 1:
The patent performs preliminary actions by organizing point cloud data into lattice distributions and identifying regions of interest before conducting detailed feature extraction. This preliminary structuring and filtering reduces the volume of data requiring intensive processing, thereby maintaining high detection precision while significantly reducing overall processing time.
Solution Approach 2:
The patent segments the feature extraction process into multiple stages with decreasing resolution and increasing focus. Early stages process data at lower resolution to identify general regions, while later stages concentrate computational effort on specific areas of interest. This segmented approach maintains high detection precision for critical obstacles while reducing total processing time through progressive refinement.
Data Source
AI summary
The disclosure provides a method for detecting an obstacle, an electronic device, and a storage medium, relates to the field of artificial intelligence, and in particular, to the field of automatic driving. A specific implementation solution includes: acquiring point cloud data collected by a lidar sensor during detection of an obstacle; converting the point cloud data into an obstacle feature map, where the obstacle feature map includes a to-be-detected image feature, and the to-be-detected image feature includes an obstacle feature; performing feature extraction on the to-be-detected image feature in the obstacle feature map to obtain the obstacle feature, increasing a weight value in the obstacle feature map of the obstacle feature during the feature extraction, and monitoring the feature extraction based on monitoring data generated by point cloud distribution data corresponding to the point cloud data; and determining a target obstacle according to the obstacle feature.