search

TheCodez / dynamic-occupancy-grid-map

Implementation of "A Random Finite Set Approach for Dynamic Occupancy Grid Maps with Real-Time Application"
MIT License
276 stars 39 forks source link

Implementation differing from paper and dogma? #58

Closed cbachhuber closed 4 years ago

cbachhuber commented 4 years ago

For the calculation of the normalization component of an unassociated measurement (mu_UA) in calc_norm_unassoc() you are using persistent_occ_mass (rho_p) divided by occ_mass.

In the paper, equation 63 states mu_UA = rho_p / pred_occ_mass. In equation 63, the pred_occ_mass is the sum of weights originally predicted into the cell, while the occ_mass in your implementation is the result after dempster's rule of combination of the predicted mass and the measured mass. Mitkina's python implementation does it as in the paper.

Is there a reason for you doing that differently?

TheCodez commented 4 years ago

Good catch! Thank you 👍 Fixed in https://github.com/TheCodez/dynamic-occupancy-grid-map/commit/010301cd5f5a1b414082ffb9f495bd2f6d492786

Given that we only have laser measurements with an association probability of 1 and no radar measurements which would have varying pA, this has no impact on the current implementation as we only have associated particles.

cbachhuber commented 4 years ago

Great, that was quick! Thanks :)