All Projects → seemk → FastHumanDetection

seemk / FastHumanDetection

Licence: MIT license
Human detection in depth images using histograms of oriented gradients

Programming Languages

c
50402 projects - #5 most used programming language
C++
36643 projects - #6 most used programming language

Fast human detection in depth images

An implementation of Fast Human Detection for Indoor Mobile Robots Using Depth Images for Kinect V2 depth images.

Differences from the paper:

  • Kinect V2 over V1
  • Region planarity checks are disabled by default
  • A neural net is used for classification instead of SVM
  • Stratified sampling is used over random sampling for point cloud construction

Building

Standard CMake build.

Generate the Makefiles:

$ mdkr build
$ cd build
$ cmake ..

Or MSVC project files:

$ mdkr build
$ cd build
$ cmake -G "Visual Studio 12 Win64" -DCMAKE_PREFIX_PATH=KINECT_SDK_DIR ..

KINECT_SDK_DIR is usually C:\Program Files\Microsoft SDKs\Kinect\v2.0_1409

Run make or build the MSVC projects.

Training a dataset

fhd_ui can be used to create a training set from depth images.

"open database" selects a Sqlite DB of depth images. Clicking on a candidate (marking it green) sets it as a positive candidate (human). Pressing space commits the candidates to the database, where selected (green) marks a human and unselected candidate marks a negative candidate. X advances to the next frame

After creating the training set, a classifier can be trained under the Training tab.

Training UI snapshot

Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].