Assembling de novo a fragment with Velvet

[MariaP]

I am trying to assemble a short (~5k-10k) region from paired-end data. First I extracted aligned reads for that region in sam format. Then I run velvet, and it produced a set of non-overlapping contigs with adistance between them about 100-200 bp. When I look at the region in IGV I see a full coverage of region with only several short (10-20 bp) areas where coverage drops to 4. Otherwise it is always >20. I can not understand why velvet is not able to reconstruct a full sequence. I used -31 with velevth, and tried -exp_cov 0 -cov_cutoff 1, it didn't change the result.

Any help would be appreciated.