Confidence structure with variance of inliers.
Our strategy for uncertainty representation is to give classification confidence scalars for classifications and confidence intervals and covariance for coordinate estimates. The classification confidence scalars are straightforward, a scalar in the [0,1] interval. For obstacles, lanes and lane edges we give the main classification confidence and also a sub-classification confidence. This is to provide access to good information for major classes that matter and also for detailed sub-classification.
For coordinates and other values we provide confidence and covariance which allow us to handle both outliers (the small but non-zero number of cases where estimates are very wrong) and the properties of the inlier distribution. We essentially provide the covariance matrix of the inlier distribution and the confidence and corresponding scaling of the confidence ellipsoid. This allows to easily test whether any new value belong to the confidence interval for example by checking that x' (covariance)^-1 x <= threshold'. The confidence (inlier ratio) in this case represents the amount of inliers within this threshold.
Definition at line 249 of file MatrixTypes.h.
Data Fields | |
| float32_t | confidence |
| Inlier ratio [0,1]. More... | |
| float32_t | threshold |
| Inlier threshold. More... | |
| float32_t | variance |
| Variance of inliers. More... | |
| float32_t dwConfidence1f::confidence |
Inlier ratio [0,1].
Definition at line 251 of file MatrixTypes.h.
| float32_t dwConfidence1f::threshold |
Inlier threshold.
Definition at line 252 of file MatrixTypes.h.
| float32_t dwConfidence1f::variance |
Variance of inliers.
Definition at line 253 of file MatrixTypes.h.