Module 4: Computer Vision & Object Detection

Welcome to Module 4: Computer Vision & Object Detection. In this module, we bridge neural networks to raw camera streams—from 2D spatial convolution kernels that extract edges and textures to real-time object detection models like YOLO that output labeled bounding boxes for robot navigation.

Concepts in this Module

• Concept 01: 2D Spatial Convolutions & Feature Maps
  • The Everyday Problem: How does a robot camera extract vertical tape lines, circular targets, and game piece edges from raw RGB pixel grids?
  • Code & Math: 3×3 convolution kernels, element-wise sliding window products, Sobel edge filters, and feature maps.
• Concept 02: Object Detection, Anchor Boxes & IoU
  • The Everyday Problem: How does an autonomous vision model (like YOLO) draw bounding boxes around game pieces and filter out duplicate overlapping boxes?
  • Code & Math: Bounding box coordinates [x, y, w, h], Intersection over Union (IoU = Area_intersection / Area_union), and Non-Maximum Suppression (NMS).