Hybrid intelligent vision-based car-like vehicle backing systems design

Abstract

This paper is integrated with discrete wavelet transformation (DWT), self-organizing map (SOM) neural network and fuzzy logic control methods to approach the intelligent vision-based car-like vehicle backing system. All the presented procedures are implemented as software simulations and are synthesized as hardware to describe their superiorities in our experimental platform. In a word, this study has been implemented as a car-like vehicle consists of a field programmable gate array (FPGA) chip, a CMOS image sensor, a microprocessor, a flash memory module, and servo motors. From our experimental results, this paper not only illustrates the results with computer simulation, but also demonstrates the practical car parking achievements in a real environment.

Introduction

It is a nonlinear and an unstable mathematical modeling problem to steer the car-like or truck into the right positions. It is hard to describe the behavior of the discussed car-like backing problems using the traditional control theories and mathematical models. Therefore, how to back a car or truck with one or more additional controller into destined targets is known as a knotty puzzle. For solving the previous questions, the objective of this paper is to design a well-defined fuzzy control system to autonomously back the car-like vehicle into the correct parking area. Forsake the traditional control theories in motion trained and active-based sensor detected schemes, the proposed modern fuzzy controller with an intelligent vision-based SOM neural network training process is considered for approaching the performed well-driving mechanism to automatically park the four-wheel car-like plant into defined targets. Since fuzzy set was introduced by Zadeh (1965), the fuzzy inference system with adjustable fuzzy rules can be regarded as a soft decision mechanism to guide the direction of the operating systems. Recently, fuzzy inference theory (Baturone et al., 2004, Fraichard and Garnier, 2001, Kong and Kosko, 1992, Tanaka and Sano, 1994, Zimic and Mraz, 2006) has been the most common used way to design a car-like vehicle fuzzy system. Numerous researches including vehicle tracking or parallel parking platform are also developed by the fuzzy logic controller. Such advantage of the fuzzy logic theory is straightforward to model the behavior of the discussed car parking space with suitable linguistic rules. In addition, the established car-like vehicle parking systems with fuzzy theory can easily add human’s experiences into the control databases. When the fuzzy rules are pre-trained well, its low computing complexity and easy hardware implementations are also established the adaptive car-like vehicles in this article.

Nowadays, many of the related studies deal with the car-like vehicle backing problems are shown in the style of computer simulations. Several researches use the active distance sensor (e.g. infrared and ultrasonic sensors) to detect the specific parking space, and then backing the car into the correct area with its specified designed methods (Li et al., 2006, Yamamoto et al., 2005, Zhao and Collins, 2005). With rapidly increasing hardware design technology, several researches not only put emphasis on the control methods validation but also complete the appropriate vehicle implementation in a real environment (Altafini et al., 2001, Chiu et al., 2005, Cuesta et al., 2004, Li et al., 2003). Due to the coordinates (i.e. size, distance and angle) of parking space are hard to be obtained, it is more difficult to implement the car-like vehicle backing system in a real environment. Therefore, its derivative problem is how to get the required measurements and estimations through the entire parking space. In sensor-detected methods of our discussed car parking system, the novel machine vision is selected to estimate vehicle’s position for moving the car toward the correct target (Chao, Ho, Lin, & Li, 2005). In this article, the built vision-based scheme is taken for extracting the necessary position information. The developed image processing methods are first proposed to generate the captured image in advance. And then, the captured images from CMOS sensor will be transformed into the image profiling information in the vector form. Without any active sensors, it can guide the controlled cars to achieve a proper parking position only with such trained image profiling information.

The efficient self-organizing map (SOM) is a famous artificial neural-network, which makes a nonlinear mapping from higher dimensional data set into lower numbers of clustering vectors (Halkidi, Batistakis, & Vazirgiannis, 2001). The SOM is trained iteratively and the weight vectors are regulated properly, the nodes will gradually approach into clusters. In the discussed motion training subjects, the powerful SOM neural networks are applied in advance to recognize some available position information from image training patterns. In this article, the SOM neural network is proposed to classify the image training data to detect the correct distances and angles from the captured image training data. Based on the previous collected message, the appropriate fuzzy control system will send a proper control signal to self-regulate directions of the vehicle to approach the intelligent vision-based car-like vehicle backing system.