# Covariance intersection

Covariance intersection is an algorithm for combining two or more estimates of state variables in a Kalman filter when the correlation between them is unknown.

## Specification

Items of information a and b are known and are to be fused into information item c. We know a and b have mean/covariance ${\hat {a}}$ , $A$ and ${\hat {b}}$ , $B$ , but the cross correlation is not known. The covariance intersection update gives mean and covariance for c as

$C^{-1}=\omega A^{-1}+(1-\omega )B^{-1}\,,$ ${\hat {c}}=C(\omega A^{-1}{\hat {a}}+(1-\omega )B^{-1}{\hat {b}})\,.$ where ω is computed to minimize a selected norm, e.g., logdet or trace. While it is necessary to solve an optimization problem for higher dimensions, closed-form solutions exist for lower dimensions. CI can be used in place of the conventional Kalman update equations to ensure that the resulting estimate is conservative, regardless of the correlation between the two estimates, with covariance strictly non-increasing according to the chosen measure. The use of a fixed measure is necessary for rigor to ensure that a sequence of updates does not cause the filtered covariance to increase.

According to a recent survey paper  and, the covariance intersection has the following advantages:

1. The identification and computation of the cross covariances are completely avoided.
2. It yields a consistent fused estimate, and thus a non-divergent filter is obtained.
3. The accuracy of the fused estimate outperforms each local one.
4. It gives a common upper bound of actual estimation error variances, which has robustness with respect to unknown correlations.

These advantages have been demonstrated in the case of simultaneous localization and mapping (SLAM) involving over a million map features/beacons.

## Development

### Pre-covariance intersection

It is widely believed that the unknown correlations ubiquitously exist in a diverse range of multi-sensor fusion problems. Neglecting the effect of unknown correlations can result in grave consequence of performance deterioration, and even divergence. As such, it has attracted and sustained attention from researchers for decades. However, owing to its intricate, unknown nature, it is not easy to come up with a satisfied scheme to address the fusion problems with unknown correlations. If we ignore the correlations directly, which is the Naive fusion, it may lead to divergence of the filter. To compensate this kind of divergence, a normal sub-optimal approach is to increase the system noise artificially. However, this heuristic requires substantial certain expertise and compromises the integrity of the Kalman filter framework.