Recursive Object Detection Filter for Bright-Target Interference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Track Before Detect methods in imaging systems face implementation challenges and failure modes, particularly when bright targets enter the field of view, requiring extensive tuning and being fragile to various failure scenarios.
Innovation Solution
A recursive object detection filter that operates in the probability domain, using probability tensors to track and detect objects by generating and updating probability tensors based on observations, avoiding traditional Bayesian filtering weaknesses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If Track Before Detect method is implemented with traditional Bayesian filtering, then detection capability is provided, but the system becomes fragile to failure modes and requires extensive tuning
Solution Approach 1:
The patent transforms the detection problem from the traditional Bayesian filtering domain to a probability domain using moment-based parameterization. By representing target states through moments (mean, variance, skewness, kurtosis) rather than full probability distributions, the system achieves robust detection while reducing implementation complexity and eliminating fragility to failure modes.
Solution Approach 2:
The patent replaces the mechanical Bayesian filtering system with a probability moment-based detection system. This substitution eliminates the need for complex tuning parameters and ad-hoc assumptions inherent in traditional Bayesian filtering, while providing more reliable detection performance through moment-based statistical characterization.
2Reliability
If traditional Bayesian filtering is used for Track Before Detect, then state estimation is provided, but the system fails when bright targets enter the field of view
Solution Approach 1:
The patent uses higher-order moments (skewness, kurtosis) to characterize the probability distribution of target signals. This parameter change enables the system to distinguish between bright targets and background clutter by capturing the shape characteristics of the signal distribution, thereby eliminating failure when bright targets enter the field of view.
3Measurement precision
If extensive tuning is applied to Track Before Detect implementation, then detection performance is improved, but the system requires application-specific tuning and remains fragile
Solution Approach 1:
The probability moment-based detection system automatically adapts to different applications and environments without requiring manual tuning. The moment parameters are computed directly from the observed data, allowing the system to self-adjust to varying conditions while maintaining high detection precision across different scenarios.
4Reliability
If probability domain approach is used instead of Bayesian filtering, then failure modes are eliminated, but computational complexity increases
Solution Approach 1:
The patent segments the probability distribution characterization into discrete moment parameters (mean, variance, skewness, kurtosis) rather than maintaining the full continuous probability distribution. This segmentation enables reliable detection through moment-based comparisons while significantly reducing computational power requirements compared to full Bayesian filtering.
Data Source
AI summary
Systems and methods are provided for detecting and tracking objects. A first set of observations for a plurality of locations includes a probability that a value associated with the location is consistent with local background. A second set of observations includes a probability that the value associated with the location is not consistent with local background. A first probability tensor representing probabilities for each of a plurality of states for each of the plurality of locations is generated from the first and second sets of observations. The first probability tensor is updated according to a second probability tensor representing the probabilities for each of the plurality of states in a previous time step. It is determined that a target is present in the region of interest when a posterior probability meets a threshold value.