Pixel Clustering for Autonomous Vehicle Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous vehicle systems face challenges in accurately determining the physical parameters of large and articulated vehicles, leading to inaccuracies in object data and vehicle control movements.
Innovation Solution
The system employs pixel clustering techniques to generate detection boxes for objects in the environment, averaging detection box parameters for pixel clusters to improve accuracy and reduce overlap, thereby enhancing the representation of large and articulated vehicles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single large bounding box is used to represent a large vehicle, then the device complexity is reduced, but the measurement precision of object boundaries deteriorates
Solution Approach 1:
The patent segments the detection task by dividing the image into multiple pixel clusters, where each cluster represents a portion of the large vehicle. Instead of using a single bounding box, the system creates multiple overlapping detection boxes from different pixel clusters, thereby improving boundary accuracy while maintaining manageable system complexity through modular processing.
2Measurement precision
If multiple separate bounding boxes are used for different parts of an articulated vehicle, then the measurement precision of object boundaries is improved, but the device complexity increases
Solution Approach 1:
The patent merges multiple detection boxes by averaging their parameters (center coordinates, dimensions, orientation) to generate a final consolidated detection box. This combining approach maintains high measurement precision from multiple clusters while reducing device complexity by producing a single unified output box for vehicle control systems.
Solution Approach 2:
The detection system is designed to handle both simple vehicles and complex articulated vehicles using the same pixel clustering and averaging methodology. The universal approach works for any vehicle type by adapting the number and distribution of pixel clusters, thereby improving boundary accuracy without proportionally increasing system complexity.
3Measurement precision
If detection boxes are generated for each pixel, then the measurement precision of object parameters is improved, but the productivity of the detection system deteriorates
Solution Approach 1:
Instead of processing each pixel individually, the patent segments pixels into clusters based on spatial proximity and similarity metrics. This segmentation reduces the number of detection boxes from potentially thousands of individual pixel boxes to a manageable number of cluster-based boxes, thereby improving processing speed while maintaining detection accuracy through clustered parameter averaging.
Solution Approach 2:
The patent combines multiple pixel detection results into averaged detection box parameters for each pixel cluster. By merging individual pixel measurements into cluster averages, the system achieves both high measurement precision (through multiple samples) and improved productivity (through reduced output volume for downstream processing).
Data Source
AI summary
Techniques for determining detection boxes representing objects in an environment using pixel clustering are disclosed herein. Autonomous vehicle sensors can capture data in an environment that may include separate objects, such as large and/or articulated vehicles. In an example, the data can include a plurality of pixels that are associated with a large object. A vehicle computing system can generate a detection box for each pixel of the plurality of pixels and can determine at least one averaged detection box representing the large object. In an example, the vehicle computing system can control the vehicle based in part on the at least one averaged detection boxes.