multi object feature detection

feature detection

A framework for feature detection. Framework supports (1)[ORB, SIFT, and SURF] with both (2)[FLANN and brute-force] nearest neigbor matching with a final pruning algorithm of (3)[crosscheck or Lowe's ratio test]. Use detect_organized.py for basic feature detection and display.py to see results. Look at detect_organized_t.py under /tests for an example

object detection

This project's main goal is to detect multiple objects. The method is composed of:

Image object segmentation

We use selective search to find potential objects

Basic cutoff

Basic cutoff of potential objects that do not have enough features to be meaningful

Matchbox reduction

Matchbox reduction using homography matricies to reduce object bounding box using homography transforms from master image to respective matchboxes

Homography angle cutoff

Using angle measurement of homography quadrilaterals to prune matchboxes that have multiple of desired object within

NMS

Non-Maximum supression to prune redundant/overlapping matchboxes

If you want to see an example of all of this in motion, check out tests/roi_trailmix_t.ipynb

cv2... more info later