Nasrabadi Abbas ¹, Vaezi Mahdi ²

Affiliations
1 Mechatronics Engineering Department, Computer and Electrical Engineering Faculty, Hakim Sabzevari University, Sabzevar, IR
2 Science and Research Branch of Azad University, Computer and Mechatronics Faculty, Tehran, IR

Abstract

Background: The dramatic increase in the traffic needs effective management that leads to the creation of intelligent transportation systems (ITS) or smart roads. Method: In this paper, a novel and quickly executable neural networks based method is presented to decide the configuration of an adaptive filter and choose among many possible segments of image. Findings: The trained network can be put in practice easily, a number of test image are presented to the software, and the program calculates the multiplication of each scene’s entropy and histogram average and based on that the trained radial basis network predicts the suitable filtering parameter. Result show the outcome of the network trial on three examples. It’s obviously because that the current input has values which are relatively close to the values put at the training stage. Applications/Improvements: A lot of research can be committed about enhanced filtering and masking ideas. A human operator is needed to approve the output’s readiness to be segmented.

Keywords

Color Filtering, Image Entropy, Machine Vision, Neural Networks, Online Road Detection

Full Text

   

Nasrabadi Abbas ¹, Vaezi Mahdi ²

Affiliations
1 Mechatronics Engineering Department, Computer and Electrical Engineering Faculty, Hakim Sabzevari University, Sabzevar, IR
2 Science and Research Branch of Azad University, Computer and Mechatronics Faculty. Tehran, IR

Abstract

Background: Images get segmented and labeled and contained objects on images are identified through many methods yet dismal work only available about the exact position extraction of object while it is very important in the future generations of fully autonomous systems which would indeed be tasked to identify and manipulate obstacles or various subjects without any human supervisions. Method: In this paper a method to apply neural networks algorithms to solve this issue without attending complicated calculations is introduced. Findings: The most important result of this work is that the fact that trained and designed network result in known transfer functions between the input and output vectors which can be easily emulated by the embedded computers which may not be able to perform far more complicated calculations of camera calibration matrixes. This is a major advantages for the robotic systems developers. Applications/Improvements: It seems that method has had quite an acceptable turnover but it should be considered that the test environment has been similar and at the same scale as the training environment.

Keywords

Camera, Images, Neural Networks, Position Extraction, Robotic

Full Text