IMM Tracker Model Selection Using Object and Road Context
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous vehicle systems face inefficiencies in model selection for object tracking, leading to increased processing requirements and potential tracking performance degradation due to the use of a large number of irrelevant models, which can delay reaction time and accuracy.
Innovation Solution
A system and method for model selection with an interacting multiple model (IMM) tracker that adaptively maintains a narrowed list of optimal models based on context information, including object and road attributes, dynamically updating the model probabilities and transition matrix to ensure accurate and efficient tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a large number of models are used for each specific type of object and attribute in the IMM tracker, then the tracking coverage and model applicability are improved, but the processing requirements and reaction time are worsened
Solution Approach 1:
The patent segments the large model set into multiple subsets based on object attributes (e.g., pedestrian, vehicle, cyclist) and motion characteristics. Each subset contains only the models relevant to that specific object type, allowing the system to process fewer models while maintaining comprehensive coverage across all object types.
Solution Approach 2:
The system dynamically selects which model subset to use based on the detected object type and context. Rather than processing all models simultaneously, the system adapts the active model set in real-time according to the specific tracking scenario, optimizing processing speed while maintaining model applicability.
2Adaptability or versatility
If a large number of models are used for each specific type of object and attribute in the IMM tracker, then the tracking coverage and model applicability are improved, but the processing requirements and computational load are worsened
Solution Approach 1:
The patent segments the large model set into multiple subsets based on object attributes (e.g., pedestrian, vehicle, cyclist) and motion characteristics. Each subset contains only the models relevant to that specific object type, allowing the system to process fewer models while maintaining comprehensive coverage across all object types.
Solution Approach 2:
The system performs preliminary classification of objects based on detected attributes before model selection. This preliminary action identifies the object type and predicts the most likely motion patterns, allowing the system to pre-select the relevant model subset before actual tracking begins, thereby reducing computational complexity.
3Reliability
If the model list is not narrowed down, then all possible motion types are covered, but the processing time increases and reaction time is delayed
Solution Approach 1:
The system dynamically selects which model subset to use based on the detected object type and context. Rather than processing all models simultaneously, the system adapts the active model set in real-time according to the specific tracking scenario, optimizing processing speed while maintaining model applicability.
Solution Approach 2:
The system performs preliminary classification of objects based on detected attributes before model selection. This preliminary action identifies the object type and predicts the most likely motion patterns, allowing the system to pre-select the relevant model subset before actual tracking begins, thereby reducing computational complexity.
4Productivity
If context information is used to narrow down the model list, then processing efficiency is improved, but the system complexity for context analysis increases
Solution Approach 1:
The system performs preliminary classification of objects based on detected attributes before model selection. This preliminary action identifies the object type and predicts the most likely motion patterns, allowing the system to pre-select the relevant model subset before actual tracking begins, thereby reducing computational complexity.
Solution Approach 2:
The context analysis module serves multiple functions: it classifies object types, predicts motion patterns, selects relevant model subsets, and updates transition probabilities. By making this single module multi-functional, the system avoids adding separate complex subsystems while still achieving efficient model selection.
Data Source
AI summary
A system for model selection with an interacting multiple model (IMM) tracker is provided. The system includes sensors associated with a vehicle that detect an object and object attributes, and includes a database that stores the object attributes, road attributes, and models capable of being assigned to the object as representative of motion type of the object. The system includes a processing device that limits the models to a shortened list based on selection of only models relevant to the object having the object attributes and the road attributes, assigns a model probability to each of the models representative of an estimation of a correct determination of relevance of each of the models, sorts the shortened model list from highest to lowest model probability, and assigns selected model(s) from the shortened model list having the highest model probability and/or a model probability greater than a predefined value to the object.