Data Science is essentially about doing things in a smarter way. So, with the help of data and tech, we develop algorithms/ rules that automate certain processes and in turn, save time. Depending on the services or products we use, we will get to experience the different applications of data science.
Recently I came across a use case of facial recognition that I thought was pretty interesting. I participated in the Allstate Hot Chocolate 15K Seattle run on 3 March and the race photos section made use of facial recognition algorithms! Of course, there's room for improvement as photos with a 90-93% confidence did not actually feature me.
A logical workflow I have in mind on how this is being done:
1) Recognising digits: Identify photos with bib number that belongs to me (they have it in their database of bib number and runner's name). However, we need to note that there are many photos with the bib number blocked from sight (as seen below) by my hands/ the object I'm carrying.
2) Collecting data on faces and attire of runners: Information on runners' faces and attire can be now linked to runner's name and bib number from photos with bib number that are obvious (i.e. not hidden/ blocked).
3) Recognising faces and attire: Identify photos with hidden bib numbers based on runners' faces and attire.
While I'm not sure if this is already being done (I'm guessing not), one way I could think of that can help in the accuracy of tagging would be identifying based on running postures as well. But that would require a lot more photo orientations of individual runners. They have also included a selfie option and that would improve the accuracy as well since users are providing accurate and clear information for step 2 in the process flow above. I was actually pretty surprised to see photos with my bib number blocked showing up in the results.