Integration of whole body motion and hand gesture tracking of astronauts to ERAS(European MaRs Analogue Station for Advanced Technologies Integration) virtual station. Skeleton tracking based feature extraction methods will be used for tracking whole body movements and hand gestures, which will have a visible representation in terms of the astronaut avatar moving in the virtual ERAS Station environment.
Benefits to ERAS
“By failing to prepare, you are preparing to fail.” ― Benjamin Franklin
It will help astronauts in getting familiar with the their habitat/station, the procedures to enter/leave it, the communication with other astronauts and rovers, etc. Thus preparing themselves by getting a before hand training in the Virtual environment will boost their confidence and will reduce chances of failures to great extent, ultimately resulting in increase in the success rate of the mission.
An idea of implementing integration of full body and hand gesture tracking mechanism is proposed after having a thorough discussion with the ERAS community. The method proposed use 3D skeleton tracking technique using a depth camera known as a Kinect sensor(Kinect Xbox 360 in this case) with the ability to approximate human poses to be captured, reconstructed and displayed 3D skeleton in the virtual scene using OPENNI, NITE Primesense and Blender game engine. The proposed technique will perform the bone joint movement detections in real time with correct position tracking and display a 3D skeleton in a virtual environment with abilities to control 3D character movements. The idea here is to dig deeper into skeleton tracking features to track whole body movements and hand gesture capture. The software should also maintain long-term robustness and quality of tracker. It is also important that the code should be less complex and more efficient. It should have more automated behavior and minimum or no boilerplate code. It should also follow the standard coding style set by the IMS(Italian Mars Society) coding guidelines.
The other important feature of the tracker software should be that, it should be sustainable long-term in order to support further future improvements. In other words, the codes and tests must be easy to modify when the core tracker code changes, to minimize the time needed to fix the code and tests after architectural changes are performed to the tracker software. This feature would allow the developers to be more confident of refactoring changes in the software itself. Following are the details of the project and the proposed plan of action.
REQUIREMENTS DURING DEVELOPMENT
- Hardware Requirements
- Kinect Sensor(Kinect Xbox 360)
- A modern PC/Laptop
- Software Requirements
- OpenNI/NITE library
- Blender game engine
- Tango server
- Python 2.7.x
- Python Unit-testing framework
- Vim (IDE)
THE OUTLINE OF WORK PLAN
Skeleton Tracking will be done using Kinect sensor and OpenNI/NITE framework. Kinect sensor will generates a depth map in real time, where each pixel corresponds to an estimate of the distance between the Kinect sensor and the closest object in the scene at that pixel’s location. Based on this map, application will be developed to accurately track different parts of the human body in three dimensions.
OpenNI allows the applications to be used independently of the specific middleware and therefore allows further developing codes to interface directly with OpenNI while using the functionality from NITE Primesense Middleware. The main purpose of NITE Primesense Middleware is an image processing, which allows for both hand-point tracking and skeleton tracking. Tracking of whole skeleton can be done using this technique however main focus of the project will be on developing framework for full body motion and hand gesture tracking which can be later integrated with ERAS Virtual station. The following flow chart gives a pictorial view of working steps.
Basically, The whole work is divided into three phases :
- Phase I : Skeleton Tracking
- Phase II : Integrating tracker with Tango server and prototype development of a glue object
- Phase III : Displaying 3D Skeleton in 3D virtual scene
Phase I : Skeleton Tracking
Under this phase comes tracking of full body movements and hand gesture capturing. RGB and depth stream data are taken from the Kinect sensor and is passed to PSDK(Prime Sensor Development Kit) for skeleton calibration.
Skeleton Calibration : Calibration is done to gain control over the controlling device.
Skeleton calibration can be done :
- Manually, or
Manual Calibration :
- For manual calibration user is require to stand in front of Kinect with his whole body visible and has to stand with both hands in air(‘psi’ position) for few seconds. This process might take 10 seconds or more depending upon the position of Kinect sensor.
Automatic Calibration :
- It enable NITE to start tracking user without requiring a calibration pose. It also helps to create skeleton shortly after user enters the scene. Although skeleton appears immediately but auto-calibration takes several seconds to settle at accurate measurements. Initially skeleton might be noisy and less accurate but once auto-calibration determines stable measurements the skeleton output becomes smooth and accurate.
However, Analyzing cons and limitation of both method. In the proposed application, I will be giving option to the user to choose among two given calibration method. Considering the fact that a user can go out of view only if the training session is interrupted. So we will ask the user (that will occupy always the same VR station) to do a manual calibration at the beginning of the week and then an automatic recalibration can happen every time a simulation restart in the same training rotation.
Skeleton Tracking : Once calibration is done OpenNI/NITE will start the algorithm for tracking the user’s skeleton. If the person goes out of the frame but comes in really quick, the tracking continues. However, if the person stays out of the frame for too long, Kinect recognizes that person as a new user once she/he comes back, and the calibration needs to be done again. Once advantage which we get here is that Kinect doesn’t require to see the whole body if the tracking is configured as the upper-body only.
output : NITE APIs will return the positions and orientations of the skeleton joints.
Phase II : Integrating tracker with Tango server and prototype development of a glue object’
As whatever we are doing here must be ready for supporting multi-player (a crew of 4/6 astronaut)so there will be 4/6 Kinect sensors and 4/6 computers supporting each a virtual station. The application must be able to populate each astronaut environment with the avatars of all crew members. The idea is that Tango will provide around the skeleton data of all crew members for cross visualization. Skeleton data obtained from each instances of tracker will be published to Tango server as tango parameters. A prototype will be developed for changing reference frame of the tracked data from NITE framework to blender reference frame and send it to blender framework for further processing. It is called a glue object since, it acts as a interface between the NITE and blender framework. Since blender has support for python bindings, this glue object will be created via Python.
Phase III : Displaying 3D Skeleton in 3D virtual scene
Under this step work will be done in-order to get all skeleton(s) data from glue object and is transferred to blender framework, where 3D skeleton will be displayed in the 3D virtual scene driven by blender game engine. Basically it provides a simulation of user in virtual environment. The idea here is that 3D skeleton inside the virtual environment will mimic the same gestures/behavior which is performed by user in real world.
An application that tracks full body movement and hand gesture for effective control of astronaut’s avatar movement with following features.
- application will detect the movement and display the user’s skeleton in 3D virtual environment in real time and the positions of the joints are presented accurately
- It can detect many users’ movements simultaneously
- Bones and joints can be displayed in 3D model in different colors with the name of user on top of head joint
- It can display the video of RGB and depth during the user movement
- Users can interact with 3D virtual scene with rotation and zoom functions while user can also see avatar in a variety of perspectives
- It can display 3D virtual environment in a variety of formats(3DS and OBJ). Also, virtual environment can be adjusted without interpretation of the motion tracking
- Proper automated test support for the application with automated unit test for each module.
- Proper documentation on the work for developers and users
To view more detailed application checkout this link – https://wiki.mozilla.org/Abhishek/IMS_Gsoc2014Proposal