Depth-Based Region-of-Interest (ROI) Selection
This is the ROI selection method used in the paper Detecting Humans in RGB-D Data with CNNs. It reduces the search space in an image and produces a set of square proposals. It was particularly designed for human upper body detection, but it can be modified to suit object detection as well.
git clone https://github.com/KaiyangZhou/ROI_Selection.git
cd
to the ROI_Selection directory and run matlab
(if you have installed)build()
to generate mex filesdemo()
to see what happensimg.mat
contains a color image and an aligned depth image (unit in meters), which are obtained from RGBD people dataset.
Core functions are enclosed in functions/roi_selection.m
. For detailed descriptions, please refer to our paper.
@inproceedings{kzhou2017humandetect,
title = {Detecting Humans in RGB-D Data with CNNs},
author = {Kaiyang Zhou and Adeline Paiement and Majid Mirmehdi},
year = {2017},
month = {2},
booktitle = {IAPR Conference on Machine Vision Applications (MVA2017)},
}