Autonomous Vehicle Object Tracking Using Nearest Anchor Points
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face inaccuracies in determining the characteristics of external objects due to the use of projected middle points, which can lead to improper estimations and predictions, especially when the size or shape of the object changes as more data is collected.
Innovation Solution
The implementation of 'near point' adjustment techniques, where the autonomous vehicle identifies and uses the nearest point on an object to determine characteristics, such as velocity and acceleration, rather than projected middle points, and couples virtual representations with measured representations to reduce discrepancies between different object representations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If projected middle points are used to determine object characteristics, then initial estimations can be made with limited data, but measurement precision deteriorates when object size or shape changes as more data is collected
Solution Approach 1:
The patent applies dynamics by transitioning from static middle point calculations to dynamic nearest point tracking. The system continuously updates the reference point on the tracked object to maintain it as the nearest point to the autonomous vehicle, rather than using a fixed middle point. This dynamic adjustment ensures that the reference point remains consistent even as the object's perceived size and shape change with additional sensor data collection.
Solution Approach 2:
The patent changes the parameter used for reference point selection from 'middle point of bounding box' to 'nearest point to vehicle'. This parameter change fundamentally resolves the inconsistency that occurs when bounding box dimensions change. By selecting the nearest point based on actual spatial relationship rather than geometric center, the system maintains measurement precision regardless of how the object's estimated size evolves.
2Ease of operation
If middle points of detected objects are used for tracking, then object characteristics can be determined, but inaccuracies occur when the autonomous vehicle adjusts its understanding of object size or shape
Solution Approach 1:
The system implements dynamic nearest point tracking where the reference point on the tracked object is continuously updated to remain the point nearest to the autonomous vehicle. This dynamic approach replaces the static middle point method, ensuring that the reference point selection adapts to changing object perceptions while maintaining tracking simplicity through automated point selection.
Solution Approach 2:
The patent creates a virtual representation (copy) of the tracked object that includes identified nearest points. This virtual model allows the system to track characteristics without being affected by uncertainties in the physical object's exact boundaries. The copying approach enables reliable tracking by working with a stabilized digital representation rather than directly processing changing sensor data.
3Speed
If projected middle points are used, then velocity and acceleration can be calculated from position changes, but improper jumps occur in characteristic determinations when object shape changes
Solution Approach 1:
The patent implements dynamic nearest point tracking where the reference point is continuously updated to remain the nearest point to the vehicle. This dynamic selection eliminates the 'jumps' that occur with middle point methods when bounding boxes change size, because the nearest point selection is based on actual spatial relationships that change smoothly as the vehicle and objects move relative to each other.
Solution Approach 2:
The patent replaces the mechanical/geometric middle point calculation system with a spatial relationship-based nearest point selection system. Instead of calculating the center of a bounding box (which is sensitive to size changes), the system identifies the point on the object nearest to the vehicle based on actual distance measurements, providing more stable and accurate velocity and acceleration calculations.
Data Source
AI summary
An autonomous vehicle comprises a sensor and one or more processors. The one or more processors can be configured to receive a set of object location data from the sensor, the set of object location data indicating an object; generate a measured representation of the object based on the set of object location data; execute a tracking protocol using data of the measured representation of the object as input to generate a virtual representation of the object, the virtual representation of the object comprising a virtual point that is nearest to a defined location of the autonomous vehicle; and translate the virtual point of the virtual representation of the object to a measured point of the measured representation of the object.