Counting objects in a "busy" image?

I am trying to write a MATLAB script to count the number of bottles visible in a still image. They have a consistent size, shape, and colour; but variable orientation, and are in amongst other objects (occasionally overlapping).

I have attempted to train a Fast-RCNN detector and count the number of detections above a certain threshold, but the results have been inaccurate due to false positives, and due to individual bottles being detected by multiple bounding boxes.

Is there a different approach which is more appropriate, or a change to my current method which would work better? (Ideally one which would be able to be repurposed for a different object)

load('bottles\detector.mat')
inputs = dir('bottles\inputs');
threshold = 0.9;
outputs = zeros(length(inputs),1);
for i = 3:length(inputs)
    img = imread(['bottles\inputs\',inputs(i).name]);
    [boxes,score,labels] = detect(detector,img,'NumStrongestRegions',Inf,'SelectStrongest',true);
    score(score<threshold) = [];
    numBottles = length(score);
    imwrite(img,['bottles\outputs\',num2str(numBottles),'_',num2str(i),'.png']);
end