Vision Based Mobile Robot Navigation System

Transcription

1 International Journal of Control Science and Engineering 2012, 2(4): DOI: /j.control Vision Based Mobile Robot Navigation Syste M. Saifizi *, D. Hazry, Rudzuan M.Nor School of Mechatronic Engineering, University Malaysia Perlis, Perlis, 02600, Malaysia Abstract The purpose of the vision syste is to recognize the circle as landark and identify the distance and orientation in a series of reference target iage autoatically fro the closed-range navigation. In order to identify position in an indoor environent, a obile robot requires attention to characteristics of a target which is to allow obile robots to ake inference in which the distance and orientation of the target and a robot. In this study, fuzzy logic is applied to generate target trajectory oveent with the inforation extracted fro vision syste such as the distance and the orientation of target. The fuzzy logic controller (FLC) which coputes the required speed and angular speed needed by the two otors to drives the robot for target trajectory. Keywords Control, Fuzzy Logic, Navigation, Differential Drive, Mobile Robot, Vision Syste 1. Introduction The navigation syste plays very an iportant role and challenging copetence for obile robot. Navigation of obile robot includes a variety of theories and technologies such odoetry technique, ultrasonic apping, and vision syste. In the application of navigation, it consists of two areas: global navigation and local navigation[1]. Global navigation such as GPS (Global Positioning Syste) and INS (inertial Navigation Syste) is often used in various open areas. Meanwhile, local navigation such as vision techniques are very effective base in the closed range navigation. There are any applications in indoor environent vehicles and obile robots. Navigation using vision approach has any tasks, including target atching, target identification and others[1]. Practically, the use of optical sensors and iage processing are the ajor factors affecting the accuracy of navigation[1]. In navigation application, a obile robot ust interpret its sensors data to extract environent inforation, with which the obile robot can deterine its position[2]. After obile robot can localize its position, it ust decide how to act to achieve its purpose. Lastly, the obile robot ust control its drive syste to achieve the desired path[2][3]. The increent distance of the robot s oveent sensed with wheel encoders is integrated to calculate position. Because easureent errors fro encoders are integrated, the position error is accuulated over tie. Using additional sensors can help to reduce the cuulative errors. There are any error sources of differential encoder * Corresponding author: saifizi@uniap.edu.y (M. Saifizi) Published online at Copyright 2012 Scientific & Acadeic Publishing. All Rights Reserved syste fro environental factors, such as isalignent of the wheels, uncertainty of wheel diaeter, variation in the contact point between wheel and floor and slipping[3][4]. 2. Vision Sensing This work presents an approach using coputer vision which applied feature atching technique after the iage has been proceeds by Canny edge detector. Features like circular arking, the distance between two vertical lines used by the fuzzy syste to analyze whether it has relation with the door being analyzed to design an expert syste to detect rectangular shaped door. In this study, vision syste is applied to extract the inforation such as the distance of target and the orientation of target Canny Edge Detector In this study, the value of threshold 0.95 for Canny edge detector was chose for both low threshold and high threshold. This process provides a continuous edge and eliinates edge the streaking significant edge[5][6] Circle Detection The circle detection technique used in this study is a features atching technique. This technique has two steps, features extraction and representation between the features. Feature extraction is a process of extracting high-level geoetric features of the curves obtained by curve extraction process. The curves and segentation of the curves are used to create circular features. A curve is a set of edge points that are connected to for a continuous contour. Curves usually represent a boundary of the part in the iage. The curve extraction process consists of three steps of finding seed point curves, tracing the curve, and refining the curves.

2 84 M. Saifizi et al.: Vision Based Mobile Robot Navigation Syste 2.3. Distance and Orientation of Circle By using a vision sensor to calculate the target distance (D t ) and circle orientation (θc) as shown in Equation (1) and (2) that refer in [7]. By knowing the diaeter and circle center coordination of the circle, the distance and orientation of circle can be calculated. Therefore, the inforation distance fro the target obile robot can be known through the iage circle diaeter. To align the target in the world coordinate frae is to use the actual object size and the coputation of object position. The set-up for this test is shown below in Fig.1. The caera is placed directly in front of the target circle. Figure 2. Differential steering syste of the obile robot Figure 1. The projection of the circle onto the caera iage plane Essentially, by knowing the diaeter of the circle and the calibration of the caera object distance can be easured by the length the iage circle diaeter. The distance between the vision sensor and the landark is inversely proportional, k to the length the iage circle diaeter. Diaeter circle, x is used to calculate the distance door-to-obile robots. The calculation used is present in following equation: 1 Dt k (1) ' x n x c (2) c center The Equation (2) was used for the angle orientation of the landark. The Circle landark orientation θ c by vision sensor can be calculated using the x-axis pixel position of the center point is calculated at the circle arking. The center circle coordinate, x center is used to calculate the angle of the circle position in the grabbing iage using n and c which are constants derived fro the caera calibration. 3. Mobile Robot Control Syste The posture provides inforation about how the obile robot oves with respect to the floor. The differential steering syste the obile robot is shown above. As shown in Fig.2, (x, y): reference frae (X, Y): robot coordinates fae R: position of robot in (x, y) θ: orientation angle of robot in (x, y) r: wheel radius 2d: distance between wheels u R/L : speed of right and left wheels respectively v: speed of robot ω: angular speed of robot and R : angular speed of right and left wheels L respectively The speed relative to the ground of the right wheel and the left wheel can be expressed as follows, respectively. R ur (3) R t L ul L (4) t The speed and the angular speed of a obile robot are related to the wheel speeds, and it can be expressed as follows: ( ur ul) v 2 r ( R L) 2 t 1 ( ur ul ) 2d r ( R L) (6) 2d t 3.1. Fuzzy Logic Controller The aster controller is a fuzzy logic controller (FLC) which coputes the required speed and angular speed needed by the two otors to drives the robot for target trajectory.

3 International Journal of Control Science and Engineering 2012, 2(4): A block diagra of the fuzzy controller is shown in Fig.3. The desired θ d and target distance D t are acquired by calculating the position of the target in the iage representing the environent detected by webca caera. The coand signal θ d and D t are transitted fro the vision syste to the fuzzy controller inside the PC. The error between the coand signal and the actual position, as well as the change in error of signal are calculated and fed into the fuzzy controller ebedded in the Data Acquisition (DAQ). Fro the Equation (1) and Equation (2), it can be seen that the differences between the speed of the right and left wheels deterines the turn speed. The fuzzy controller is designed to output pulse width odulation (PWM) signal corresponding to u R and u L to the right otor and left otor respectively to control the obile robot turn angle θ to the desired angle θd and target distance D t. inference echanis is accountable for decision aking in the control syste using estiated sensors that give the inforation of the target object [8]. The operation involved in this thesis is AND because two inputs D t and θ d are involved. Fig. 6 and Fig. 7 show the output ebership functions. Note that the output PWM signals should be selected to eet the convergence requireents. The ebership of the input and output was designed based on the training of obile robot odelling it based differential drive technique. Figure 6. Output ebership functions for the speed Figure 3. A cascade control syste for target trajectory The fuzzification procedure aps the crisp input values to the ebership values between 0 and 1. In this study, three ebership functions for target distance D t and five ebership functions desired angle θ d was used as shown in Fig. 4 and Fig. 5 illustrate the input ebership functions for D t and θ d respectively. Figure 7. Output ebership functions for the angular speed Figure 4. Input ebership functions for target distance Dt The control rules are designed based on expert knowledge and testing. For exaple, if θ d is poslarge and is D t far, then the left otor should be uch than the right otor, i.e., u L -u R should be left and speed u L +u R should be. Based on knowledge, 15 rules were obtained. Table 1 represents how to control the turn angle given the angle and distance as inputs. The input and output linguistic variables are shown in the table. Table 1. Table rules for the trajectory steering syste θ d Figure 5. Input ebership functions for desired angle θ d The rule base stores the rules relating to the knowledge input output relationship of the offered fuzzy controller. The D t far close too close Neg large Neg sall center Pos sall Pos large

4 86 M. Saifizi et al.: Vision Based Mobile Robot Navigation Syste 4. Experients and Results Experient results fro distance and orientation calculation are shown in Table 2 and Table 3. Absolute and relative errors are also presented. The relative error was calculated using as following equation: vision anual relative(%) 100 (5) anual easured Table 2. Experiental results for the calculation of the distance to a landark Manual Measured[ eter] Vision Measured[ eter] Absolute Error[ eter] Error[ %] Table 3. Experiental results for the calculation of the angle orientation to a landark Manual Measured[ ] Vision Measured[ ] Absolute Error[ ] Error[ %] Although the vision syste is only using a single webca, distance calculation of target showed quite accurately. The average absolute error was eter and the average relative error was 3.26% for distance easureent. Given the easureent angle of orientation, it also revealed quite accurately with a ean absolute error of 3.16, and the average relative error of 15.66%. Referring to[7], the coparison can be ade that Equations (1) and (2) can be used not only in the target in shape of a rectangular but can also be applied to the target in the shape of a circle to enable the calculation of the distance and orientation of objects using onocular vision. Due to the design of the landarks and the usage of single webca, the preliinary step of calibration is necessary. This calibration has to be done for the extraction of distance and orientation for an iage. The usage of only one iage akes it difficult to have extreely accurate results although atheatical geoetry is used the process is still dependent on caera calibration. Considering the obile robot otion, although low error values are present in the calculation of both distance and orientation, in the robot s oveent these values will be constantly corrected. The obile robot captures iages periodically and even if soe easureents have soe error. By providing inforation fro the vision syste, fuzzy logic onitors the status of the robot. The robot is able to navigate in the laboratory, thus fuzzy logic controls the robot to ove and position itself in the right conditions. The vision systes recognize the circle, the position of the circle and then align the robot to be in the center of the circle arking. As the perforance of an autonoous robot depend on the interaction of the robot with the real world for any eaningful conclusion to be drawn, experients have to be done to deonstrate that the robot can perfor the desired tasks. Overall perforance test of obile robot was conducted to analyzed cobination of fuzzy logic controller (FLC) rules with desired speed. The developed fuzzy logic control syste cobined with the vision syste as a sensor to extract uch inforation (target distance and target orientation) fro the iage to achieve a high level of self-adaptability is very iportant. The robot oved in different distances fro the target and stop at 1.5 eter fro the target because that is the iniu distance that circle can recognize. The Tables 4 and 5 show the perforance of obile robot in ipleentation of fuzzy logic controller. When the robot oving to the target, the results show that he average absolute error was about 0.03 eter and the average relative error was less than 3% for distance of robot to achieve it target. Given the easureent angle of orientation, it also revealed quite accurately with a ean absolute error of about 20, and the average relative error of about 22%. Table 4. The orientation of the robot when arrived at target Departure Distance [ eter] Robot Orientation [ ] Absolute Error [ ] Error [ %] Average Table 5. The distance of the robot when arrived at target Departure Distance[ e ter] Robot Distance [ eter] Absolute Error [eter] Error [ %] Average

5 International Journal of Control Science and Engineering 2012, 2(4): Conclusions Landark recognition is not an easy task but it is very iportant area in developing obile robots that is need to accoplish positioning, path planning and navigation. A robot vision syste using fuzzy logic technique to identify the object in real tie was presented. The syste is able to find and recognize the circle and then calculate the position of the robot to navigate in a laboratory. The technique used is based on iage processing and optiization in real tie. It hard to an able a obile robot navigates in indoor environent. The output is very useful for a obile robot in real tie to find and detect objects. Usually a robot vision syste is very due to coputational tie used for iage processing. Iage processing tie was iniized by the techniques used and the detection of door can be done about less than 1 second. An analysis and design of fuzzy control law for steering control of the developed nonholonoic obile robot are presented. The proposed fuzzy controller is ipleented on the developed obile robot. The syste can perfor and satisfactory results are obtained which show that the proposed fuzzy controller can achieve the desired turn angle thus it can ake the autonoous obile robot oving to the target. The control syste need PID controller to guarantee the stability of the straight conveying and turning trajectory. ACKNOWLEDGEMENTS The First of all, I would like to acknowledge the contributions of the any people who provided invaluable help and support for the copletion of this thesis. My supervisor, Assoc. Professor Dr. Hazry Desa, deserve thanks for not only guiding e with a steady hand, but also leaving e the freedo to pursue y own research interests. REFERENCES [1] Batlin and Maxi A, Mobile robot navigation using a sensor network. In: IEEE international conference on robotics and autoation, [2] Saidon M.S, D. Hazry, Nagarajan R, Paul Raj MP, Vision based tracking control of an autonoous obile robot in an indoor environent, Proceedings IEEE Control and Syste Graduate Research Colloquiu, ICSGRC 2011 (2011), Article nuber , Pages 1-6, [3] Saidon M.S, D. Hazry, Rudzuan M.N, A Differential Steering Control with Proportional Controller for an Autonoous Mobile Robot, Proceedings IEEE 7th International Colloquiu on Signal Processing and Its Applications, CSPA 2011 (2011), Article nuber , Pages 90-94, 2011 [4] D. Hazry and M. Sugisaka, Optial Paraeter Tuning in Proportional Control for the Unicycle Mobile Robot: An Experiental Study, Proceedings of the 11th International Syposiu on Artificial Life and Robotics (AROB 2006), 2006 [5] Canny, J., A Coputational Approach to Edge Detection, In: IEEE Transactions on Pattern Analysis and Machine Intelligence. Vol. 8, No. 6, pp , [6] Khalil, A., A. Aggoun, A. ELabrouk, On Edge Detector Using Local Histogra Analysis, In: SPIE Visual Counications and Iage Processing. Vol. 5150, [7] Yoon, K.-J., G.-J. Jang, S.-H. Ki, and I. S. Kweon, Fast Landark tracking and localization algorith for the obile robot self-localization, In: IFAC Workshop on Mobile Robot Technology, pages , [8] Driankov, D., H. Hellendoorn and M. Reinfrank, An Introduction to Fuzzy, Control. Springer, Berlin, 1998.