You are on page 1of 31

Person Re-identification via

Structural
Deep Metric Learning
Proposed topic

Deep Dense-Net enabled network for


Multi-Spectral image identification
Objective
• To implement a robust detection
framework for person re-identification
regardless of direction

• To implement a deep learning based


algorithm for pattern identification of
similar person tracking system
Scope of the study
• This study enables us to understand the
idea of deep learning algorithms.
• This study provides deep knowledge on
image classification techniques using
MATLAB
• To understand social needs on person re-
identification in terms of security
Abstract
• Problems in person re-identification Is discussed in the present
system.

• regardless of colour, key identification parameters objects are


required to Verify the the person.

• Data set is collected from CCTV images a pair of free


identification pattern is analysed using deep learning densenet.

• Performance measure is Analysed using accuracy with respect


to the state-of-art approach developed in the existing system.
Existing System
• In this paper, we present an effective person re-ID framework
by discriminatively learning a nonlinear deep feature mapping
from person images to low-dimensional embedding's, where
similar samples are mapped closer to each other, while
dissimilar samples are pushed farther apart.
• The proposed approach jointly learns feature representation
and distance metric in an end-to-end manner.
Proposed System
• Image re-identification regardless of color and
structure is developed here using Deep
DenseNET
• Evaluation of performance measure using the
feature extraction technique through Brisk –
MinEigen features.
• Object identification is included
Applications

• CCTV Surveillance for Secure operations

• Tracking of pedestrians

• Forensics, abnormal activity detection


Problem Statement

• Color based identification creates confusion


in similar color objects
• Fast movement of tracking images give large
span of pixel variations
Solution

• Re-identification Vector is created using BRISK-


MinEigen features
• Deen Dense-Net can analyze the patterns
iteratively
Software SPecs
• MATLAB tool
• Version 2017
• Language : MATLAB C Commands
Toolbox :
• Deep learning
• Image processing
• Neural Networks
Block diagram
METHODOLOGY
Process of implementation undergoes in 5 basic steps.
• 1. Image Acquiring.
• 2. Pre-processing
• 3. Segmentation
• 4. Feature Extraction
• 5. Classification.

CONVOLUTIONAL NEURAL NETWORK


• Convolutional Neural network is a deep learning technique which is feed forward
artificial neural networks that are applied to visual images. These are also called as
CONV NETS. These are combined set of layers that can be worked on group
functionalities.
Image acquiring

• The Primary Phase is acquiring images. After the Images

collection, the obtained images have to be prepared with a

wide range of vision. First capture the input images from

available source with uigetfile and imread mat lab functions.


Pre Processing
• The images which are collected are subjected to preprocessing. In Pre-
processing stage basic steps are image resizing and applying Gaussian
filters for a perfect input clear image for easy identification of an image.
1. Segmentation
In this stage of implementation, Pre-processed images will be segmented
digitally into various pixels. We do this segmentation for an image is to
modify its representation to have more clarity to study and analyse the
images.
Feature Extraction

• In the feature extraction process, we can


implement the effective texture operator which
labels the pixels of an image. Here we extract
the features and characteristics of Images for
easy detection of person.
Classification
• In Classification stage, Convolutional neural networks algorithm
is used for classification of images. It is a non-parametric
method which is used for both classification and regression.
• DenseNet

A DenseNet is a type of convolutional neural network that


utilises dense connections between layers, through Dense Blocks,
where we connect all layers (with matching feature-map sizes)
directly with each other.
Image acquisition and classification flow
chart
Color cooccurence method
• Color Co-occurrence Method (CCM) uses HSI pixel maps to
generate three unique Spatial Gray-level Dependence
Matrices (SGDM)

• Each sub-image was converted from RGB (red, green, blue)


to HSI (hue, saturation, intensity) color format

• The SGDM is a measure of the probability that a given pixel


at one particular gray-level will occur at a distinct distance
and orientation angle from another pixel, given that pixel
has a second particular gray-level
Neural networks
“ A neural network is a system composed of many
simple processing elements operating in parallel
whose function is determined by network structure,
connection strengths, and the processing performed
at computing elements or nodes ”.
(According to the DARPA Neural Network Study (1988, AFCEA International Press,
p. 60)
Contd..
“A neural network is a massively parallel distributed processor
that has a natural propensity for storing experiential
knowledge and making it available for use. It resembles the
brain in two respects:

1. Knowledge is acquired by the network through a learning


process.
2. Inter-neuron connection strengths known as synaptic
weights are used to store the knowledge.

[According to Haykin, S. (1994), Neural Networks: A Comprehensive Foundation, NY:


Macmillan, p. 2]
A Basic Neuron
Multilayer Feed forward Neural Network
Back propagation
• In the MFNN shown earlier the input layer of
the BP network is generally fully connected to
all nodes in the following hidden layer
• Input is generally normalized to values
between -1 and 1
• Each node in the hidden layer acts as a
summing node for all inputs as well as an
activation function
MFNN with Back propagation
• The hidden layer neuron first sums all the
connection inputs and then sends this result
to the activation function for output
generation.
• The outputs are propagated through all the
layers until final output is obtained
Mathematical equations

The governing equations are given below:

N
u   wi xi
i 1

y  f (u   )
Where x1,x2… are the input signals,
w1,w2…. the synaptic weights,
u is the activation potential of the neuron,
is the threshold,
y is the output signal of the neuron,
and f (.) is the activation function.
Back propagation
• The Back propagation algorithm is the most
important algorithm for the supervised
training of multilayer feed-forward ANNs
• The BP algorithm was originally developed
using the gradient descent algorithm to train
multi layered neural networks for performing
desired tasks
Back propagation algorithm
• BP training process begins by selecting a set of
training input vectors along with
corresponding output vectors.
• The outputs of the intermediate stages are
forward propagated until the output layer
nodes are activated.
• Actual outputs are compared with target
outputs using an error criterion.
Back propagation
• The connection weights are updated using the
gradient descent approach by back
propagating change in the network weights
from the output layer to the input layer.
• The net changes to the network will be
accomplished at the end of one training cycle.
BP network architecture used in the research

Network Architecture:
2 hidden layers with 10 processing elements each
Output layer consisting of 4 output neurons
An input layer
‘Tansig’ activation function used at all layers
Conclusion
• The research was a feasibility analysis to see
whether the techniques investigated in this
research can be implemented in future real
time applications.
• Results show a positive step in that direction.
Nevertheless, the real time system involves
some modifications and tradeoffs to make it
practical for outdoor applications

You might also like