Our team developed this project for the Fall 2022 Waterloo.AI Data Challenge, sponsored by Musashi AI. The task was to detect people in selected areas of overhead fisheye camera footage. We explored how machine learning could handle the distorted views these cameras produce in workplace settings.

Our approach combines a pretrained ResNet-101 model for extracting image features with k-means clustering to help identify people. We also explored fine-tuning and linear classifiers as alternative ways to approach the detection task.