If you are interested in transcript counts, use an appropriate tool for the task. You may map with STAR (as you did) and count with RSEM or eXpress. Even better, you could get the counts directly from an indexed transcriptome with kallisto or Salmon. These tools take into account the redundant nature of transcripts and apportion multi-mapping reads optimally using an EM algorithm.
If you are using
FPKM_count.py from RSeQC, it requires a bed file, not a gtf.