Which solution will meet these requirements with the LEAST operational overhead?
Use Amazon Rekognition.
Use a custom convolutional neural network (CNN).
Use the Amazon SageMaker Object Detection algorithm.
Use Amazon Lookout for Vision.
Explanations:
Amazon Rekognition is a fully managed service for image and video analysis, including text detection, which is ideal for extracting race IDs from images of runners with minimal operational overhead. It requires no custom model training and is easy to integrate for this specific use case.
Using a custom CNN would require significant operational overhead, such as model training, tuning, and maintenance. This approach is more complex than necessary for extracting text in this case.
Amazon SageMaker Object Detection is designed for detecting specific objects in images, not text. It would require custom configuration and is not as efficient as a text recognition service for this task.
Amazon Lookout for Vision is primarily used for detecting anomalies in images, not extracting specific text. It is not suited for the task of race ID extraction, making it an inefficient choice.
I plot that the answer is:
Use Amazon Rekognition.