Challenges

The Image recognition model had been overtrained and now detects anything rectangular object that a person is holding as a threat

Inadequate amount of time to pair Pose recognition with Knife image recognition

The model was too resource intensive and cannot be used on small IOT devices such as Arduinos and Rasberry Pis which was its original goal

Prototyping and Testing :

The Model had a 70% functional and operation success rate

The model had a 40% chance of a misprediction

Results/Conclusion :

Achieved Top 5 place.

(I did not know of these results until very late when I was randomly talking to the instructor, he told me I got top 5 out of the blue, however getting top 5 was not enough to qualify for a presentation :P)