AI Bird Watching

My birds need to be watched, but I cannot always watch them. Using the ImageAI Python library I can do this easily. With ~20 lines of code I can find if there are any birds in an image, and count how many with varying degrees of accuracy.

The setup_detector function sets up the ObjectDetection model from ImageAI. In this model I choose to load the TinyYoloV3 model, chosen because it is the most lightweight model, and I plan to run this on my Raspberry PI.

from imageai.Detection import ObjectDetection

def setup_detector():
  """
  Build the detection model
  :return: The detection model
  """
  execution_path = os.getcwd()
  detector = ObjectDetection()
  detector.setModelTypeAsTinyYOLOv3()
  detector.setModelPath(os.path.join(execution_path, "./yolo-tiny.h5"))
  detector.loadModel()

Using this model it is time to detect the birds. I only want to see the birds in the image do I set the CustomObjects to birds only. Then by feeding in my input image I get the image seen below as output.

import os

def process_picture(image_path, output_path):
  """
  Process the input image
  :param image_path: Input image path
  :param output_path: Output image path
  :return: Found objects
  """
  detector = setup_detector()
  custom_object = detector.CustomObjects(bird=True)
  detections = detector.detectObjectsFromImage(custom_objects=custom_object,
                                               input_image=image_path,
                                               output_image_path=output_path,
                                               minimum_percentage_probability=1,
                                               display_object_name=False)

  return detections

With these ~20 lines of code we have found all the birds in our image. As you can see the probability percentages are quite low, but this can be improved with a different model. The full guide including all model choices can be found on the full ImageAI Detection documentation.