Download a bed-file with snp coordinates (because otherwise like you said most coordinate will have BAF=0):

Go to UCSC tables and choose your preferred snp-table (example) and for output format choose BED.

Then make a script called mpileup2baf.pl:

#!/usr/bin/perl
use strict;
use warnings;
use Getopt::Long qw(:config pass_through no_ignore_case);

my ($min_reads) = (10);
GetOptions (
        "min-reads:s" => \$min_reads,
);

while (<>) {
        my $line = $_;
        my @columns = split("\t",$line);
        my $chr = $columns[0];
        my $start = $columns[1];
        my $end = $start + 1;
        my $num_reads = $columns[3];
        my $calls = $columns[4];
        my $id = "mpileup_number_" . $.;
        if($num_reads < $min_reads){ # not enough coverage to have good confidence in the call
                next;
        }
        my $num_ref = 0;
        while ($calls =~ /[,.]/g) { $num_ref++ }
        my $num_var = $num_reads - $num_ref;
        my $varAlleleFreq = ($num_var/$num_reads)*100;
        print("$chr\t$start\t$end\t$id\t$varAlleleFreq\n");
}

Then use samools mpileup to get all the base calls at the snp-postitions and calculate BAF with mpileup2baf.pl

samtools mpileup -q 15 -Q20 -f hg19.fa -l snp137.bed yourSample.bam | mpileup2baf.pl --min-reads 20 > yourSample_mpileup_snp137_minReads20_minMapQ_15_minBaseQ_20.bed

This skips bases with base quality lower than 20 and bases in reads with mapping quality lower than 15. Furthermore it ignores snp positions where the amout of reads that are mapped is lower than 20.Then plot with any visualization program you like. I prefer ggplot2