It's a report to demonstrate a project which implements a prototype to make Kinect camera stability and prepare a better ability of detecting, recognizing body movement such as waving, Squat, Push and pull.
Auther:
Chen-I Chang
Kent Chang
3. Motivation
It's been hours for the modern people to stay in
front of the computer for a long time and health
problems may be the case in the long run. Here
is a way we think of: the use of somatosensory
game so that modern people up activities, no
longer be restrained in front of the computer
can not only achieve the purpose of relieving
pain, helping the injured patient to recuperate
and build confidence, and can be at leisure
Enjoy entertainment.
4. Game Intro
Long, long time ago, the Earth was hit by
natural disasters and the creatures were
facing a crisis of extinction. A group of
animals saw a ship on the other side of
the sea, they felt the boat can escape
the danger, but the ship can no longer
accommodate other things, the player
needs to repel the animals, the player
can survive.
5. Architecture
Kinect: Analyze the user's joints and pass
the skeleton to the Controller.
Controller: Joint computing user aiming
point, detection push, gesture recognition.
Unity Engine: load scene, animal system,
engine, explosion effects, animation
system ... and so on
Scene: shows the score, menu, user
interface, time.
8. Basic Theory
Use the left and right rotation of the
body skeleton, tilt back and forth to
change the camera aiming
position, and detect hand
trajectory, determine gesture to
operate and launch attack ball, hit
the specified target to obtain
points.
Person-oriented position: Use the
joint of Right and Left Shoulder
and the joint of Right and Left Hip
to form X vector, use Y-vector from
Shoulder Center and Hip Center,
Cross into third vector, and face-to-
face orientation.
9. Basic Theory
Problem:
Because Kinect detection is very
accurate, if you use the body's
immediate vector as
Targeting may cause camera shake
and screen jitter problems.
Solution:
Using the Weighting Average to
smooth the input information to
improve the camera when the
problem will be fluttering aim to get
20,30 aiming point within the
detection to calculate an aiming
point and updated after each frame
to recalculate, The best result is 0.3.
11. Gesture recognition
Problem: Identify correctness
Because Kinect's built-in gesture recognition in Unity
Wrapper is not accurate, sometimes the action will be
misjudged, so we've done some real-world ways of
identifying specific actions.
Solution:
Using the joint parameters of the joint and the shoulder
of the hand, the distance and the vector information can
be obtained, and judging the massive change at a
certain time to identify the Push or Pull triggered, so that
the posture can be more accurately identified.
14. Basic Theory
Push and Pull detection:
Use Shoulder, Elbow, Hand three joints to do
detection. 1.Shoulder and Hand are between vectors
(+ -0.2, + - 0.2,> 0). 2. Shoulder and Hand distances
produce a z-axis shift for a limited time. 3.Hand's y
vector approach elbow's y vector.
15. Basic Theory
Posture recognition:
1.TopTorso, ButtonTorso composition of the vector to
determine the left and right tilt, you can determine
the squat.
2.Right and Left Hand, Elbow, Shoulder determine
the relative position.
16. Basic Theory
Posture recognition:
1.TopTorso, ButtonTorso
composition of the vector to
determine the left and right tilt,
you can determine the squat.
2.Right and Left Hand, Elbow,
Shoulder determine the relative
position.
17. Future
Hope to increase the game's
interactive, support double
game, and will increase the
battle
Combination, combined with
cell phone connection, let a
party control the launch of a
bullet, one control the walking
and emergence of animals,
and identify the action in the
picture below.