Robotic Tracking Navigation Using Adaptive Bayesian Data Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Robotic navigation systems face challenges in robust visual tracking due to issues like object and camera motion, orientation changes, illumination variations, occlusion, scale changes, and clutter, making single approaches unreliable for short-term scenarios and nearly impossible for long-term applications.
Innovation Solution
A hierarchical adaptive Bayesian data fusion method is employed, combining multiple tracking algorithms at the bounding box level using Mahalanobis and Euclidean distances as weighting factors, with Kalman Filters and auto-encoders to adaptively fuse sensor measurements and detect detector failures, enabling robust real-time tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple tracking algorithms are used to improve robustness, then tracking reliability improves, but system complexity increases
Solution Approach 1:
The patent combines multiple independent tracking algorithms (DSST, TLD, CMT) into a unified Bayesian fusion framework. Each algorithm processes sensor data independently to generate local estimates, which are then merged through adaptive Bayesian fusion to produce a single robust target location estimate. This merging approach improves tracking reliability by leveraging the strengths of multiple algorithms while managing complexity through a structured fusion architecture.
Solution Approach 2:
The tracking system is segmented into independent detector modules, each implementing a specific tracking algorithm. These modules operate autonomously to generate local estimates, which are then combined by a fusion center. This segmentation allows each module to be optimized independently while the overall system benefits from diverse tracking approaches, resolving the contradiction between reliability and complexity.
2Measurement precision
If more sensors are added to improve tracking accuracy, then measurement precision improves, but device complexity increases
Solution Approach 1:
The Bayesian fusion framework serves as a universal processing architecture that can accommodate multiple sensor types and tracking algorithms through a common interface. The system processes bounding box measurements from various detectors uniformly, adapting to different sensor inputs without requiring separate processing pipelines. This multi-functionality improves measurement precision by integrating diverse sensor data while avoiding the complexity of dedicated processing systems for each sensor type.
3Measurement precision
If adaptive Bayesian fusion with Mahalanobis distance is used to reduce uncertainty, then tracking precision improves, but computational complexity increases
Solution Approach 1:
The system uses Mahalanobis distance calculation as a parameter change mechanism to dynamically weight local estimates based on their statistical properties. By computing the Mahalanobis distance between the global estimate and local estimates, the system adapts the weighting factors to reduce uncertainty in a statistically optimal way. This approach improves tracking precision by accounting for correlations in measurement errors while managing computational complexity through efficient distance calculations.
4Reliability
If detector failure detection is implemented to improve reliability, then system robustness improves, but processing time increases
Solution Approach 1:
The system implements feedback through auto-encoders that continuously monitor detector outputs for failures. The auto-encoders learn normal detector behavior and generate reconstruction errors that serve as failure indicators. This feedback mechanism improves system robustness by detecting and responding to detector failures in real-time while managing processing time through efficient neural network-based anomaly detection.
Data Source
AI summary
Disclosed are systems and methods of sensor fusion for exemplary use with robotic navigation control. Systems and methods include providing local estimates of a target location from a plurality of expert modules that process sensor data. The local estimates are weighted based upon a Mahalanobis distance from an expected estimated value and based upon a Euclidean distance between the local estimates. The local estimates are fused in a Bayesian fusion center based upon the weight given to each of the local estimates.