Image edge detection using SVM regression model for UAV autonomous navigation

Resumo

Currently, unmanned aerial vehicle (UAV) has been used for many applications. Few applications
include agriculture, engineering, monitoring, rescue, entertainment, and others. One research topic involving UAV
is related to autonomous navigation. A standard procedure for this process to UAV is to combine inertial sensor
information with the Global Navigation Satellite System (GNSS) signal. However, some factors can interfere with
the GNSS signal associated with natural phenomena or malicious attacks (jamming or spoofing). One alternative
to overcome using the GNSS signal is to apply an image processing approach based on matching UAV images
and georeferenced images. There is a great computational effort on this approach for computing the image edge
extraction. A Support Vector Machine (SVM) regression model, also known as Support Vector Regression - SVR,
is employed for edge detection to reduce computational load and processing time. Our proposal consists of three
general steps; first: pre-processing of the images, where frames of 3x3 pixels were obtained for characterizing
edge or non-edge patterns; second, SVR models were trained, where the predictors were normalized; and finally,
an i.i.d. (independently and identically distributed) test set was used to predict SVR respective responses. The
better-performing model was acquired using the Gaussian kernel function compared to two other kernel functions
(linear and polynomial). Its generalization error is that the out-of-sample mean-squared error (MSE) was 18 times
less than the Linear kernel MSE error. The success rate was 99.98% of accuracy.