Object height using Kinect

Question

For example, I am standing in-front of my Kinect. The Kinect can identify the joints, and it will expose them as a data structure. Till this point I am clear.

So, can we define the height as the difference between Head joint - ((LeftAnkle + RightAnkle)/2)?

I have tried trigonometric formulas, but there are two problems I am facing. One is identifying the person in the view. The second one is identifying the exact positions of Top of head and bottom of foot.

I have tried the point cloud, but got lost in how to generate the point cloud specific to a person. I mean without including the background objects.

Please suggest some ideas about how I can calculate the height of a person using the Kinect?

Inherit

Which api are you using to access the Kinect? – Freshman

I am using Windows Kinect API. – Inherit

Take a look to this answer too – Cosmonautics

Possible duplicate of Measuring person's height using Kinect – Cosmonautics

Answer 1

You can convert the Head Joint into global coordinate system. There is no need to do any math. The y coordinate in global coordinate will be his height.

All you need to do is check what pixel the head joint is and convert the pixel + depth informations into word coordinate space in mm.

I don't know what API you are using, but if it's being capable to segment a human and return his joint's, probably you are using OpenNI/NITE or Microsoft SDK. Both of them have a function that converts a pixel + depth coordinate into a x,y,z in mm. I don't know exactly what are the functions but their names would be something like : depth_to_mm, or disparity_to_mm. You need to check both documentations to find it, or you can do it by yourself. This site have informations on how to convert depth to mm: http://nicolas.burrus.name/index.php/Research/KinectCalibration

Osmen

I have tried this approach. I am able to get the Depth information almost correctly. i can say almost 1 or 2 inches difference, which is ok. But when i check the Y coordinate value, it's not at all matching with the original height. even not the half value.there is huge difference between the y-coordinate value and Original value. – Inherit

It's something like, My height is 5.5ft. When take the y-ordinates it's saying 1.2ft. Any idea, what could be the error. – Inherit

Are you converting from mm to ft? – Osmen

Answer 2

I have extracted the two points - Head and Left foot (or Right Foot), then i found the euclidean distance between these points gave the distance with 4 inch variation. My test results are satisfactory, so we are using this approach as temporary work around.

Inherit

Answer 3

An old question but i found a very nice explanation and example here. It also explains that height isnt mearly a function of the head and ankle points, but instead a function of the following line segments:

• Head - ShoulderCenter
• ShoulderCenter - Spine
• Spine - HipCenter
• HipCenter - KneeLeft or KneeRight
• KneeLeft / KneeRight - AnkleLeft / AnkleRight
• AnkleLeft / AnkleRight - FootLeft / FootRight

Practitioner

Answer 4

Here is a formula for Kinect SDK 2.0. Full project available at https://github.com/jhealy/kinect2/tree/master/020-FaceNSkin_HowTallAmI ....

using System;
using Microsoft.Kinect;

// Skeleton is now Bones
public enum BodyHeightMeasurementSystem
{
    Meters = 0, Imperial = 1
}

public static class BodyHeightExtension
{
// change this to change the way values are returned, by default everything is meters
public static BodyHeightMeasurementSystem MeasurementSystem = BodyHeightMeasurementSystem.Meters;

/// <summary>
/// Get Height of a body in CM
/// </summary>
/// <param name="TargetBody">used for extension method purposes - uses should not see</param>
/// <returns>
/// positive value: height in meters
/// -1.0 : null body passed in
/// -2.0 : body not tracked, no height available
/// </returns>
public static double Height( this Body TargetBody )
{
    if ( TargetBody == null ) return -1.0;
    if (TargetBody.IsTracked == false) return -2.0;

    const double HEAD_DIVERGENCE = 0.1;

    Joint _head = TargetBody.Joints[JointType.Head];
    Joint _neck = TargetBody.Joints[JointType.Neck];

    // var spine = skeleton.Joints[JointType.Spine]; // ?
    Joint _spine = TargetBody.Joints[JointType.SpineShoulder];
    // var waist = skeleton.Joints[JointType.HipCenter];  // ?
    // jeh: spinemid is ignored
    Joint _waist = TargetBody.Joints[JointType.SpineBase];
    Joint _hipLeft = TargetBody.Joints[JointType.HipLeft];
    Joint _hipRight = TargetBody.Joints[JointType.HipRight];
    Joint _kneeLeft = TargetBody.Joints[JointType.KneeLeft];
    Joint _kneeRight = TargetBody.Joints[JointType.KneeRight];
    Joint _ankleLeft = TargetBody.Joints[JointType.AnkleLeft];
    Joint _ankleRight = TargetBody.Joints[JointType.AnkleRight];
    Joint _footLeft = TargetBody.Joints[JointType.FootLeft];
    Joint _footRight = TargetBody.Joints[JointType.FootRight];

    // Find which leg is tracked more accurately.
    int legLeftTrackedJoints = NumberOfTrackedJoints(_hipLeft, _kneeLeft, _ankleLeft, _footLeft);
    int legRightTrackedJoints = NumberOfTrackedJoints(_hipRight, _kneeRight, _ankleRight, _footRight);

    double legLength = legLeftTrackedJoints > legRightTrackedJoints ? Length(_hipLeft, _kneeLeft, _ankleLeft, _footLeft) 
        : Length(_hipRight, _kneeRight, _ankleRight, _footRight);

    // default is meters.  adjust if imperial to feet
    double _retval = Length(_head, _neck, _spine, _waist) + legLength + HEAD_DIVERGENCE;
    if (MeasurementSystem == BodyHeightMeasurementSystem.Imperial) _retval = MetricHelpers.MetersToFeet(_retval);

    return _retval;
}

/// <summary>
/// Returns the upper height of the specified skeleton (head to waist). Useful whenever Kinect provides a way to track seated users.
/// </summary>
/// <param name="skeleton">The specified user skeleton.</param>
/// <returns>The upper height of the skeleton in meters.</returns>
public static double UpperHeight( this Body TargetBody )
{
    Joint _head = TargetBody.Joints[JointType.Head];
    // used to be ShoulderCenter. Think its SpineMid now
    Joint _neck = TargetBody.Joints[JointType.SpineMid];
    // .Spine is now .SpineShoulder
    Joint _spine = TargetBody.Joints[JointType.SpineShoulder];
    // HipCenter is now SpineBase
    Joint _waist = TargetBody.Joints[JointType.SpineBase];

    return Length(_head, _neck, _spine, _waist);
}

/// <summary>
/// Returns the length of the segment defined by the specified joints.
/// </summary>
/// <param name="p1">The first joint (start of the segment).</param>
/// <param name="p2">The second joint (end of the segment).</param>
/// <returns>The length of the segment in meters.</returns>
public static double Length(Joint p1, Joint p2)
{
    return Math.Sqrt(
        Math.Pow(p1.Position.X - p2.Position.X, 2) +
        Math.Pow(p1.Position.Y - p2.Position.Y, 2) +
        Math.Pow(p1.Position.Z - p2.Position.Z, 2));
}

/// <summary>
/// Returns the length of the segments defined by the specified joints.
/// </summary>
/// <param name="joints">A collection of two or more joints.</param>
/// <returns>The length of all the segments in meters.</returns>
public static double Length(params Joint[] joints)
{
    double length = 0;

    for (int index = 0; index < joints.Length - 1; index++)
    {
        length += Length(joints[index], joints[index + 1]);
    }

    return length;
}

/// <summary>
/// Given a collection of joints, calculates the number of the joints that are tracked accurately.
/// </summary>
/// <param name="joints">A collection of joints.</param>
/// <returns>The number of the accurately tracked joints.</returns>
public static int NumberOfTrackedJoints(params Joint[] joints)
{
    int trackedJoints = 0;
    foreach (var joint in joints)
    {
        // if (joint.TrackingState == JointTrackingState.Tracked)
        if ( joint.TrackingState== TrackingState.Tracked )
        {
            trackedJoints++;
        }
    }
    return trackedJoints;
}

/// <summary>
/// Scales the specified joint according to the specified dimensions.
/// </summary>
/// <param name="joint">The joint to scale.</param>
/// <param name="width">Width.</param>
/// <param name="height">Height.</param>
/// <param name="MaxX">Maximum X.</param>
/// <param name="MaxY">Maximum Y.</param>
/// <returns>The scaled version of the joint.</returns>
public static Joint ScaleTo(Joint joint, int width, int height, float MaxX, float MaxY)
{
    // SkeletonPoint position = new SkeletonPoint()
    Microsoft.Kinect.CameraSpacePoint position = new Microsoft.Kinect.CameraSpacePoint()
    {
        X = Scale(width, MaxX, joint.Position.X),
        Y = Scale(height, MaxY, -joint.Position.Y),
        Z = joint.Position.Z
    };
    joint.Position = position;
    return joint;
}

/// <summary>
/// Scales the specified joint according to the specified dimensions.
/// </summary>
/// <param name="joint">The joint to scale.</param>
/// <param name="width">Width.</param>
/// <param name="height">Height.</param>
/// <returns>The scaled version of the joint.</returns>
public static Joint ScaleTo(Joint joint, int width, int height)
{
    return ScaleTo(joint, width, height, 1.0f, 1.0f);
}

/// <summary>
/// Returns the scaled value of the specified position.
/// </summary>
/// <param name="maxPixel">Width or height.</param>
/// <param name="maxBody">Border (X or Y).</param>
/// <param name="position">Original position (X or Y).</param>
/// <returns>The scaled value of the specified position.</returns>
private static float Scale(int maxPixel, float maxBody, float position)
{
    float value = ((((maxPixel / maxBody ) / 2) * position) + (maxPixel / 2));

    if (value > maxPixel)
    {
        return maxPixel;
    }

    if (value < 0)
    {
        return 0;
    }

    return value;
}

}

Rowlandson