There are lots of ways to do this. I think the easiest solution is to use knowledge of known exon junctions.
1). Providing TopHat with a .gtf file should produce a junctions output file (technically, it already produces this file, but I think it should be empty unless you provide a reference list of transcript locations)
2) Use a software to predict splicing events. This will give you a prioritized list (in addition to providing counts for all relevant splicing junctions). MATS is my favorite tool for this, and MISO is another popular option.
MATS: http://rnaseq-mats.sourceforge.net/
MISO: http://genes.mit.edu/burgelab/miso/
3) If it is relevant, there are also gene fusions programs. I have a slight preference for chimerascan, but I have tried all of the following programs:
TopHat-fusion: http://tophat.cbcb.umd.edu/fusion_tutorial.html
chimerascan: https://code.google.com/p/chimerascan/
deFuse: http://compbio.bccrc.ca/software/defuse/
I'm sure there are also options for generically splicing the .bed file for split reads, but I would typcially focus on looking for software that also assists with the downstream analysis (for whatever specific application I am interested in). So, I don't really have recommendations on this end.
Is it possible to count spliced read over intron from intron.bed?