Head Detection using Image Processing for a Human Detection System

1
Head Detection using Image Processing for a Human
Detection System

• By Cheng Mun Wai

2
Project Description

• Realistic Talking Head Model.
• 3D Head responds to visual stimulus.
• Visual Input.
• Movement of users head.

3
Proposed Solution

• A System that
• Tracks movement of the users head.
• Real-Time.
• Send signals to the 3D Head.
• Allows 3D Head to correctly face the user.

4
Implementation.

• Digital Video Camera.
• Separate machine.
• Connected to 3D Head (via TCP/IP).
• Sends data on users head position by request.
• Real-Time processing on a PC.

5
System Architecture
6
Head Detection Program

• Uses DirectX Media API.
• Image processing algorithms
• Normal Distribution
• Colour Ratio model (chromatic colours)
• 3D RGB colour space lookup table
• Blob Detection
• 3-thread Backbone.
• Replies requests from 3D-Head, providing users
  head co-ordinates.

7
Skin Colour Sampling (1)

• RGB values of skin colour are collected.
• Samples converted to Chromatic Colour Format.
• Strips away brightness component.
• Calculates the ratio of Red and Green channels.
• Red Ratio R / (R G B)
• Green Ratio G / (R G B)
• Blue Ratio B / (R G B)

8
Skin Colour Sampling (2)

• Red Ratio 255 / (255 0 0) 1
• Green Ratio 0 / (255 0 0) 0
• Blue Ratio 0 / (255 0 0) 0
• Total 1
• Red Ratio 255 / (255 255 0) 0.5
• Green Ratio 255 / (255 0 0) 0.5
• Blue Ratio 0 / (255 0 0) 0
• Total 1
• Red Ratio 83 / (83 143 173) 0.21
• Green Ratio 143 / (83 143 173) 0.36
• Blue Ratio 173 / (83 143 173) 0.43
• Total 1

R 255 G 0 B 0
R 255 G 255 B 0
R 83 G 143 B 173
9
Skin Colour Sampling (3)
10
Skin Colour Sampling (4)

• Red Ratio 0.36
• Green Ratio 0.34
• Blue Ratio 0.30
• Total 1.00
• Red Ratio 0.37
• Green Ratio 0.34
• Blue Ratio 0.29
• Total 1.00
• Red Ratio 0.38
• Green Ratio 0.33
• Blue Ratio 0.29
• Total 1.00

R 250 G 235 B 202
Hue 29 Sat. 211 Lum. 226
R 215 G 197 B 162
Hue 28 Sat. 102 Lum. 189
R 142 G 122 B 106
Hue 19 Sat. 37 Lum. 124
11
Skin Colour Sampling (5)

• The colour ratio values for all skin coloured
  pixels are Normally Distributed.

Red Ratio values Mean 0.35 Std. Dev. 2.67
Everything between 3 SDs from mean is considered
skin colour. (99.7 of entire collection).
12
Skin Colour Sampling (6)

• Every RGB colour is converted to Colour Ratio
  format.
• To qualify as skin colour
• Both Red and Green ratio ONLY, must be within 3
  SDs of the their respective mean.
• Result
• Of all possible colours, which RGB values are
  skin coloured.

13
Skin Colour Recognition (1)
The RGB of each pixel is checked if it is skin
coloured.
14
Skin Colour Recognition (2)
Blob Detection used to isolate groups of skin
coloured pixels. Largest group is considered the
head
Centre co-ordinate of largest blob sent to 3D
Head.
15
Thank You.