Pre-processing for GATK pipeling: BWA necessary for SNPs discovery in genomic data?

Hello,
I am working with DNA capture, with paired-end sequences, that I plan to run into GATK pipeline for SNPs and InDel Discovery in genome data.
On the website, GATK suggests to start the pre-processing of the data with bwa mem for mapping the data to a reference. I did that, but I could never found a correct way (and working with my data) to extract only the reads that mapped to my reference uniquely (I have looked for days in forums and else...). Would you have a suggestion for this matter?
Otherwise, would you recommend me to map my data with Bowtie2 instead? To your opinion, do you think I may regret it later in the GATK workflow of analyses?
Any help is more than welcome
Thanks!