Obstacle Detection Using Vehicle-Centric Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current obstacle detection technologies face high computational costs, high equipment costs, and low accuracy due to the complexity of processing detection data from multiple sensors, which limits the ability to accurately estimate the position and velocity of objects around a moving vehicle.
Innovation Solution
A method for detecting obstacles using an electronic device that transforms sensor data into a vehicle-centric coordinate system, applies clustering algorithms to group detection points, and aligns bounding boxes to determine target objects, thereby reducing computational costs and improving accuracy by combining overlapping boxes and selecting appropriate models for different object categories.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If optimization methods or neural networks are used for object detection and estimation, then detection accuracy is improved, but computational cost increases
Solution Approach 1:
The patent uses simple geometric models (bounding boxes) and basic mathematical operations instead of complex neural networks. The solution replaces expensive computational methods with cheaper, simpler algorithms that can be executed efficiently on resource-constrained devices while still providing adequate detection accuracy for obstacle tracking.
Solution Approach 2:
The patent extracts only the essential features needed for obstacle detection (position coordinates from detection points) and discards unnecessary complex processing. By focusing on extracting minimal required information and using simple averaging for position estimation, the system achieves acceptable accuracy with significantly reduced computational overhead.
2Ease of manufacture
If radar sensing solution is used for object tracking, then equipment cost is reduced, but detection accuracy deteriorates
Solution Approach 1:
The patent combines multiple detection points from the radar sensor to form a unified object representation through clustering and bounding box operations. By merging multiple data points and using geometric calculations to determine object center positions, the system compensates for the inherent limitations of single-sensor radar data and improves detection accuracy without requiring additional expensive sensors.
3Device complexity
If only point cloud is used as measurement point, then device complexity is reduced, but measurement precision deteriorates
Solution Approach 1:
The patent performs preliminary clustering and bounding box alignment operations on the detection points before final position estimation. By pre-processing the point cloud data to group points into coherent object representations and align bounding boxes with object orientations, the system prepares the data in advance to enable accurate center position calculation using simple geometric operations.
Solution Approach 2:
The patent transitions from working with raw 3D point cloud coordinates to a 2D bounding box representation in a simplified coordinate system. By projecting and aligning bounding boxes, then calculating centers in this reduced dimensional space, the system achieves accurate position estimation with simpler mathematics compared to direct 3D point cloud analysis.
Data Source
AI summary
A method for detecting obstacles implemented in an electronic device includes obtaining detection point data from at least one sensor, the detection point data comprising a first position coordinate of each of a plurality of detection points; transforming the first position coordinate of each of the plurality of detection points to a second position coordinate of each of the plurality of detection points relative to an ego vehicle; determining at least one target object according to a plurality of second position coordinates; calculating target object data of the at least one target object according to a second position coordinate of each of the plurality of detection points relative to the ego vehicle; tracking the at least one target object according to the target object data.