Among　the　widely　applied　localization　algorithms,　Monte　Carlo　localization　(MCL)　stands　out　due　to　its　nonlinear　nature.　MCL　utilizes　occupancy　grid　maps　for　particle　filtering-based　localization.　However,　the　current　methods　used　to　construct　occupancy　grid　maps　often　discretize　the　environment　into　grid　cells,　limiting　their　ability　to　represent　the　environment　accurately.　The　optimization　trade-off　challenges　the　performance　of　MCL.　In　this　paper,　an　improved　MCL　method　(MCL-PD)　is　proposed　based　on　probability　density　maps　to　address　this　issue.　The　continuous　probability　density　maps　are　constructed　using　lidar　data,　allowing　the　map　to　dynamically　change　during　the　localization　process.　Additionally,　the　map＇s　scoring　range　and　morphology　are　adjusted　based　on　particle　weights　to　amplify　the　gradient　of　the　particle　weight,　ensuring　the　selection　of　the　optimal　particle.　Experimental　results　demonstrate　that　our　proposed　MCL-PD　method　outperforms　classical　algorithms　such　as　MCL,　Normal　Distribution　Transform　(NDT),　and　Iterative　Closest　Point　(ICP)　in　terms　of　localization　accuracy.　©　2023　IEEE.