Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd
Upload
42 views

Testing of Different Machine Learning Algorithm For Detecting

The document discusses using machine learning algorithms like convolutional neural networks to detect damage on car exteriors by analyzing images. It describes collecting a dataset of damaged car images and using a Mask R-CNN model that extracts features, proposes regions of interest, classifies objects, and segments pixels to identify damage in images. The document outlines the steps taken to develop an offline system for damage detection using a Mask R-CNN model hosted on a local server.

Uploaded by

John Jairo Sanabria Sarmiento
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
42 views

Testing of Different Machine Learning Algorithm For Detecting

The document discusses using machine learning algorithms like convolutional neural networks to detect damage on car exteriors by analyzing images. It describes collecting a dataset of damaged car images and using a Mask R-CNN model that extracts features, proposes regions of interest, classifies objects, and segments pixels to identify damage in images. The document outlines the steps taken to develop an offline system for damage detection using a Mask R-CNN model hosted on a local server.

Uploaded by

John Jairo Sanabria Sarmiento
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 18

Testing of different machine learning

algorithms for detecting damage on car


external body
Subramanya Somanadha Ayyappa swami, Kambala Gangadhara
Matrikel Nr:642331
Need for a company to detect damages on auto?

Shortens Facilitates
customer global
service customer
time support

Unlimited Saving
access to customer
cloud data money
Artificial neural networks
Deep neural networks
Backpropagation
• Backpropagation algorithm algorithm

• CNNs
• RNNs
• GANs…
Example: animation of backpropagation
Convolutional neural network (CNN, or ConvNet)

• CNNs are designed to recognize visual patterns directly from pixel images with
minimal pre-processing.
Terminology:
Convolutional Layer:A convolution is the simple application of a filter to an input
that results in an activation. Repeated application of the same filter to an input
results in a map of activations called a feature map, indicating the locations and
strength of a detected feature in an input, such as an image.
Pooling layer:Pooling is nothing other than down sampling of an image. The most
common pooling layer filter is of size 2x2, which discards three forth of the
activations.
Padding: Padding is a term relevant to convolutional neural networks as it refers to
the amount of pixels added to an image when it is being processed by the kernel of a
CNN.
Dataset
• I collected data for my training and evaluation through some of the sites listed
below and collected images on own from local auto shops.
https://www.google.com/search?q=car+damage+pictures&rlz
=1C1CHBF_deDE882DE882&oq=car+damage+pictures&aqs
=chrome..69i57j0i22i30l2.6258j0j7&sourceid=chrome&ie=UTF-8
https://www.shutterstock.com/search/damaged+car
https://www.gettyimages.de/fotos/damaged-car?
mediatype=photography&phrase=damaged%20car&sort=mostpopular
https://www.istockphoto.com/de/fotos/damaged-car?mediatype=photography&ph
rase=damaged%20car&sort=mostpopular
https://www.freeimages.com/search/car-damage
Mask R-CNN flowchart
The algorithm flow is the following.

• Input the image to be processed into a pre-trained ResNet50+FPN network model


to extract features and obtain corresponding feature maps.
• This feature map obtains a large number of candidate frames (i.e., the region of
interest(ROI)) through region proposal network (RPN), and then uses the softmax
classifier to perform binary classification of foreground and background, using
frame regression to obtain more accurate candidate-frame position information, and
filtering out part of the ROI by non-maximum suppression.
• The feature map and the last remaining ROI are sent to the RoIAlign layer, so that
each ROI generates a fixed-size feature map.
• Finally, the flow goes through two branches, one branch entering the fully
connected layer for object classification and frame regression, and the other
entering the full convolution network (FCN) for pixel segmentation.
Resnet structure and improved ResNetV2 structure
Steps for Offline development for Mask R-CNN damage detection
Localhost server for hosting the website:
Uploading an image for analysis from local folder
and processing it for damage analysis
Status bar indicating after post analysis
Damage detection after post processing
Helper manual to assist newbie’s with process

You might also like