Sunteți pe pagina 1din 2

Student name: Caterine Oliveira

Student ID: 1651419


University of Adelaide

UAV Project: Detection of Landing


Pads Using Image Processing
Cascade Classifier for Object Recognition Approach
Caterine Oliveira

Summary
In this project, the main objective was built a program to the identification of landing pads
to UAV`s. The program created was basically a classifier able to identify objects of
interest, which in this case was the landing pads with an X or H in their centers. For
the de development of the codes able to create, manipulate and prepare the database for
training as well as the executables for the training and testing of the classifier it was used
the OpenCV library. OpenCV is an open source library for image processing. The library
was installed on the operation system Linux and the Platform Eclipse C++ it was utilized
to build and Debug the codes for Database manipulation and testing of the classifier. The
training process was made using the executables available at the OpenCV package
through command lines at the command window. All the steps described above were
performed on the operation system Linux.
For the training process it was build a Database composed by 4534 positives images. To
create the database it was token videos of the lading pads in different locations (grass,
asphalt and clay). These videos were processed and a frame per each second was
generated and converted to JPG image. The landing pad in each image was then extracted

Student name: Caterine Oliveira


Student ID: 1651419
University of Adelaide

and saved as a positive sample. To complete the process for the creation of the positive
database, all the images were transformed to grayscale and had their histograms
equalized. A code was created to grayscale and to equalize the histogram of each single
positive sample and a file containing the path location and image size for each image was
generated.
For the training process it was also build a Database composed by 10000 negative images.
These images were obtained from internet, were transformed to grayscale and had their
histograms equalized. A code was also created to grayscale and to equalize the histogram
of each single negative sample and a file containing the path location and image size for
each image was generated.
After the creation of the samples it was used an OpenCV executable named
opencv_createsamples to generate a vector containing the information about the set of
samples.
Once created the vector of samples the cascade classifier can be trained. To perform the
training of the classifier it was utilized the executable from OpenCV named
opencv_traincascade.
After training the classifier an xml file was generated. This file has all the information
required to the classifier identify the object of interest and also the information required
to avoid false recognitions. The file was utilized to test a range of frames containing or
not the landing pads, so it was possible to evaluate the performance and robustness of the
classifier.
It was observed after the test of several frames that the classifier is able to recognize the
landing pads in most of the cases. However, in some cases the classifier is fail identifying
the object of interest and in few cases there were false recognitions.

S-ar putea să vă placă și