Systems and methods for identifying user-customized relevant individuals in an ambient image at a doorbell device are provided. Such systems and methods can include receiving user input that includes image information, using the image information to compile a custom image database containing a plurality of images that depict such relevant individuals, and storing the custom image database in local memory of the doorbell device. Then, such systems and methods can include capturing an ambient image with a camera of the doorbell device, determining whether any person depicted in the ambient image matches any of the relevant individuals by comparing the ambient image to the plurality of images at the doorbell device, and generating an alert when any person depicted in the ambient image matches any of the relevant individuals.