Amazon offers many opportunities to apply artificial intelligence to monitor its rich biodiversity. For example, it is possible to use Deep Learning models to detect Amazonian bird species that frequent residential feeders from images. We demonstrated this in a previous study, but without automating data collection. Therefore, in this extended work, we employed webcams connected to low-cost Orange Pi Zero 3 boards to automate the recording of 6 Amazonian bird species that frequent a residential feeder. Given the volume of data collected, we also trained a preliminary Faster R-CNN model with images of a newly observed species known as the Great Kiskadee and those from previous work to partially annotate the birds in the recordings. Finally, 2,200 new images were randomly extracted from the detected recordings, and 3,358 annotations were manually reviewed and adjusted to train a final Faster R-CNN model that achieved mAP of 99.45%, mean precision of 98.47% and mean recall of 99.68% considering IoU threshold at 50%. With this, we intend to keep collecting and detecting new images to build a monitoring system to study and preserve these species in the future.