Cardinality balanced multi-target multi-Bernoulli filtering using adaptive birth distributions

In random finite set based tracking algorithms, new-born targets are modeled using birth distributions. In general, these birth distributions have to cover the complete state space. In Sequential Monte Carlo (SMC) implementations, a high number of particles is required for an adequate representation of the birth model. In this contribution, a measurement driven adaptive birth distribution is proposed for the SMC and Gaussian mixture (GM) versions of the cardinality balanced multi-target multi-Bernoulli (CB-MB) filter. It is shown that a filter with adaptive birth distribution nearly achieves the performance of a filter with known birth locations. Additionally, an application of the filter to vehicle tracking using real-world sensor data is presented.