[Question]Did you treat a Det /Track state vector as a multi-dimentional distribution?

[CaptainEven]

1. As far as i know, JS/KL divergence is designed for two distribution similarity measurement, and the Det/Track state vector is only a sample not a distribution?
2. Each element of the state vector element has a different dimension and order of magnitude(or scale), is JS divergence similarity of two state vector reasonable?

[hejiawei2023]

1. The commonly adopted approach in the TBD framework is to maintain a Kalman filter for each trajectory. Under this assumption, the trajectory state is essentially assumed to follow a Gaussian distribution. Since the trajectory state initially comes from the detection state, so detection state can also be considered as a distribution.
2. Sure. Different state distributions have different dimensions and scales. To address this issue, we adopt [x, y, z, l, w, h] to measure the similarity between two distributions, while introducing a penalty term for the difference in orientation angle. You can find more details in this link: https://arxiv.org/abs/2303.01786