AWR1642: reduce the false objection detection

Hi,

   I have processed the  obj->detObj2D data by clustering (dbscan) and tracking(Kalman) algorithms,but there are still some unwanted points in the final detection. Expcept to CFAR,are there any other methods to remove unwanted detections?Or are there some features to distinguish the real and false objects?

Thanks in advance,

Regards,

Rata