Object Identification Device Using Map-Based Region Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object identification systems face high processing loads due to the need to consider multiple motion models, including uniform linear motion, left turning, and right turning, when identifying objects over time, especially in dynamic environments like roads where vehicles follow consistent paths.
Innovation Solution
An object identification device that uses map information to divide the road into regions, converting observed positions into traveling direction and transverse direction positions, and determining object identity by comparing differences within specified thresholds, thereby reducing processing complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple motion models (uniform linear motion, left turning, right turning) are considered to identify objects, then identification accuracy is improved, but processing load increases
Solution Approach 1:
The patent divides the area into multiple regions based on map information, and applies different motion models to different regions. Specifically, it uses uniform linear motion model for regions where objects are likely to move straight, and turning motion models only for regions where turns are expected. This segmentation approach maintains identification accuracy by applying appropriate motion models locally while reducing overall processing load by avoiding unnecessary motion model calculations in all regions.
2Reliability
If all motion models are calculated to continue identification when an object turns, then identification reliability is improved, but processing time increases
Solution Approach 1:
The patent dynamically selects motion models based on regional characteristics and object behavior. Instead of continuously calculating all motion models, the system adapts the motion model selection to the current situation: using uniform linear motion model when objects move straight, and switching to turning motion models only when turns are detected or expected in specific regions. This dynamic approach maintains identification reliability while significantly reducing processing time by avoiding unnecessary calculations.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
A map information storage unit (12) that stores map information, a region determination unit (14) that acquires observed position information indicating observed positions at which an object is observed from a plurality of sensors, and determines whether each observed position is in any one of regions into which an area indicated by the map information is divided, a road reference position conversion unit (15) that converts each observed position determined to be in the region of the map information, into a traveling direction position indicating a position in a direction parallel to an assumed road direction in the region, and a transverse direction position indicating a position in a direction perpendicular to the assumed road direction in the region, using the map information, and a comparison unit (17) that rearranges the observed positions in order of the traveling direction, creates pairs of front and rear observed positions in the traveling direction, calculates a difference in the traveling direction positions and a difference in the transverse direction positions between each pair of observed positions, and determines that a pair of observed positions between which the differences are within thresholds specified in respective items are derived from the same object, and determines that a pair of observed positions between which at least one of the differences is greater than the threshold are derived from different objects, are included.