Object Tracking Type Correction Using Behavior Model Likelihoods
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image detection systems face inaccuracies in estimating the position and type of objects, particularly distinguishing between pedestrians and two-wheeled vehicles, due to limitations in their deep-learning methods.
Innovation Solution
A method and apparatus that utilize a combination of free behavior and constant speed models to accurately determine the type of an object by calculating likelihoods using Mahalanobis distances and correcting object types based on threshold values, incorporating multiple detectors for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Difficulty of detecting and measuring
If deep-learning methods are used to search for object areas in images, then object detection capability is improved, but position and speed estimation accuracy deteriorates
Solution Approach 1:
The patent introduces an intermediary correction mechanism that uses behavior models (constant speed model and free behavior model) to mediate between the deep-learning detector's output and the actual object state. The Mahalanobis distance calculation serves as an intermediary metric to evaluate and correct the detector's position and speed estimates by comparing them with predicted values from behavior models, thereby improving measurement precision without sacrificing detection capability.
2Adaptability or versatility
If image sensors are used to determine object types, then detection coverage is improved, but object type classification accuracy deteriorates
Solution Approach 1:
The patent implements a feedback mechanism where the detected object type and movement characteristics are fed back into the behavior model selection process. The system continuously monitors the object's movement pattern and uses this feedback to switch between constant speed model and free behavior model, thereby improving classification accuracy. The feedback loop allows the system to learn from detection results and adjust behavior model weights dynamically.
Solution Approach 2:
The patent changes the parameter of behavior model selection based on detected object characteristics. By calculating Mahalanobis distances and comparing them with threshold values, the system dynamically adjusts which behavior model (constant speed or free behavior) is applied to each object type. This parameter change approach enables accurate differentiation between pedestrians and two-wheeled vehicles based on their distinct movement patterns.
3Measurement precision
If behavior models are used to correct object types, then object type accuracy is improved, but system complexity increases
Solution Approach 1:
The patent segments the object tracking system into distinct functional modules: deep-learning object detector, behavior model selector, Mahalanobis distance calculator, and type corrector. Each module performs a specific function, making the overall complex system manageable and maintainable. The segmentation allows independent optimization of each component while maintaining clear interfaces between them.
Solution Approach 2:
The patent creates simplified copies of behavior models (constant speed model and free behavior model) that replicate essential movement patterns without requiring complex simulations. These copied models provide sufficient accuracy for object type classification while keeping computational requirements manageable. The Mahalanobis distance calculation uses these simplified model copies to generate correction signals.
Data Source
AI summary
A method of tracking an object includes obtaining a likelihood of a free behavior model of the object and a likelihood of a constant speed model of the object using the position, the speed, and the type of the object determined at a previous time point and the position, the speed, and the type of the object determined at the current time point, and correcting the type of the object at the current time point using the likelihood of the free behavior model, the likelihood of the constant speed model, and the measured type of the object.