The use of laser range finder on a robotic platform for pipe inspection

Mechanical Systems and Signal Processing 31 (2012) 246–257

Contents lists available at SciVerse ScienceDirect

Mechanical Systems and Signal Processing journal homepage: www.elsevier.com/locate/ymssp

The use of laser range finder on a robotic platform for pipe inspection Zheng Liu n, Dennis Krys 1 Institute for Research in Construction, National Research Council Canada, 1200 Montreal Road, Ottawa, ON, Canada K1A 0R6

a r t i c l e in f o

abstract

Article history: Received 16 April 2010 Received in revised form 9 March 2012 Accepted 12 March 2012 Available online 24 March 2012

In this paper, we investigate the use of a laser range finder on a robotic platform for buried water pipe inspection. A robotic platform carrying and manipulating multiple nondestructive inspection sensors may require accurately locating robot’s body in the pipe. The laser range finder provides an accurate distance measurement, which can generate a profile of the pipe inner surface. This profile, on one hand, can be used to identify the location of the laser source and thus the robot’s body. Such information can further help the navigation of the robot. On the other hand, the anomalies presented in the profile can be detected and characterized in terms of the range measurement. The simulated and real data tests presented in this paper demonstrate the feasibility and effectiveness of incorporating the laser range finder into a robotic platform for the underground pipe inspection. Crown Copyright & 2012 Published by Elsevier Ltd. All rights reserved.

Keywords: Laser range finder Underwater vehicle Nondestructive inspection Water pipe

1. Introduction Nondestructive inspection or testing (NDI/NDT) plays an important role for the assessment of pipe performance and conditions. A pipe deterioration model usually needs multiple inputs from inspection [1,2]. Any individual NDI technique has its own limitation; therefore, multiple techniques are preferred to provide complementary information in one inspection. An autonomous robotic platform can be a carrier to manipulate multiple NDI sensors and perform the inspection of pipe. Laser range measurement is one of the NDI techniques considered in an integrated NDI system for buried water pipes. Laser sensor has been used as a profiler for advanced geometry inspection and high-resolution 3D imaging for pipeline assessment on smart pigs and various robotic platforms [3–8]. A laser sensor used for pipe inspection may have two major functions including the assisting platform’s navigation and generating interior profile of the pipe wall. Usually, the laser profiler can operate in two working modes. In the first mode, a laser ring is projected onto the internal pipe surface, inspection camera captures the stripes in an video sequence. Postprocessing will extract the laser light stripes in the video sequence and reconstruct the profile to characterize the pipe surface [9–12]. The second mode implements a direct distance measurement capable of pinpointing the geometrical and surface anomalies, although the principles behind the distance measure may be different [3,13,14]. The relative position and orientation of the robotic platform provide paramount information for vehicle navigation and inspection. The common method for tracking the position and orientation of the platform is to use a relative positioning device such as gyroscope or inclinometer [7]. However, this type of tracking is susceptible to many errors. Therefore, laser

n

Corresponding author. Tel.: þ 1 613 9933806; fax: þ 1 613 9931866. E-mail addresses: [email protected] (Z. Liu), [email protected] (D. Krys). 1 Tel.: þ 1 613 9910936; fax: þ 1 613 9931866.

0888-3270/$ - see front matter Crown Copyright & 2012 Published by Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ymssp.2012.03.006

Z. Liu, D. Krys / Mechanical Systems and Signal Processing 31 (2012) 246–257

247

Table 1 Summary of laser profiling techniques for pipe interior shape measurement. Equipment

Approach

References

Omni-directional laser and omni-directional camera Laser projector and CCD camera Structured laser Laser profiler and camera A laser pointer and range camera Laser (diode) light source and CCD camera

Light section method and structure from motion analysis 3D shape reconstruction from the 2D structured light stripe images Inner-contour reconstruction through a mathematical model Recursive Gaussian filtering and Hough transform; sensor calibration Optical triangulation and image mapping Edge detection for laser ring; neural network for defect detection

[10,28] [11] [22] [12,24] [29] [30–32]

sensor is used for the tracking purpose. In [15], a conical laser light projected on the pipe wall was acquired by a camera. The features of the image, namely centroid and shape signature, are extracted and matched with the features in a database to estimate relative position and orientation. However, the feature database has to be prepared in advance and has an impact on the estimation accuracy and performance. Nassiraei et al. used a rotating laser scanner on their robotic prototype, namely KANTARO, to detect pipe manholes and joints as the navigational landmarks [16,17]. The laser scan data were fused with image data from a fish eye camera to implement pipe fault detection [16]. In [18,19], a line laser beam projected on the internal surface of the pipe was used to detect landmarks, such as elbows and branches, by matching the line pattern. Together with the information from motor encoders and 3D orientation sensor, a 3D map for the pipeline can be reconstructed [18]. Laser profiling inspection provides information on the post-installation behavior of pipelines. In [20]. Zhuang et al. presented the potential of using a circular optical cutting method for pipe inner wall inspection. The proposed system consisted of a laser diode light source, an optical ring pattern generator, and a CCD camera. However, the detailed discussion about the inspection results was not available in their publication [20]. Duran et al. used a laser projector together with a diffuser to generate lighting profiler on the inner surface of sewer pipes [9]. A CCD camera captured the ring projections and a neural network (NN) was trained to discriminate the shape of defects. One advantage of this method is that it utilizes image intensity variations to characterize defects and is not affected by the misalignment of the center line. In [21], Sinha et al. presented the use of neuro-fuzzy algorithm to classify defects from CCTV (closed-circuit television) survey. However, a heuristic learning like NN based approach needs ‘‘good’’ training data, which might not be always available in practice. Wang et al. proposed using a distributing laser-spot method, where multiple laser beams were simultaneously projected onto pipe surface [22]. A real-time model was developed to calculate the coordinates of these laser spots and a 3D contour was built. However, this application did not target the inspection of the pipe’s inner surface. The use of laser range finder for railway tunnel inspection was also reported [23]. A summary of laser profiling techniques is given in Table 1. As stated in [7], the limitations of currently available commercial systems may generate inaccurate measurements of the pipe and negatively impact the data due to the setup issue between the laser and the camera. Ritter et al. proposed a laser scanning system with four laser triangulation modules, which allows a very precise measurement of pipe geometry [24]. A comprehensive description of the methodologies for automated sewer pipeline defect detection is presented by Guo et al. in [25]. Similar applications of laser scan were also reported by the literature [26,27,24]. In this study, we investigate the potentials of using laser range finder for purposes of robot navigation and pipe inspection. Rather than using a structured light, a laser range finder is employed to profile the surface with direct distance measurement. Such range measurement is used for both the alignment of center line and characterization of surface anomalies. This paper proposes the algorithms to position the robot’s body and detect pipe surface anomalies from the precise distance measure. The rest of the paper is organized as follows. Section 2 describes the laser range finder and algorithms used for data pre-processing, center alignment, and anomaly detection. Experimental results are presented in Section 3. This paper is summarized in the final Section 4. 2. Pipe inspection with laser range finder 2.1. Laser range finder Laser range finder uses a laser beam to determine the distance to a reflective object. It operates on the time-of-flight principle. A laser pulse in a narrow beam is sent out to the object and the returning time of the reflected pulse is recorded. An AccuRange-4000 laser range finder, which can take laser range measurements in a 3601 profile, was used in this study. One distinguished feature of this laser range finder is that the laser emitter and return signal collection lens are concentric [33]. This allows to redirect the laser light by putting a mirror in front of the laser sensor. 2.2. Data pre-processing Every measurement from the laser range finder has an amplitude reading from 0 to 255, which indicates the signal strength. If the signal strength is very low, the measurement is more likely to have a greater amount of error involved.

248

Z. Liu, D. Krys / Mechanical Systems and Signal Processing 31 (2012) 246–257

There are many factors that may cause a weak laser signal, such as the geometry, roughness, reflectivity, and color of the pipe surface. An empirical threshold value is applied to remove weak signals, which may result in unreliable measurements. In this study, 30 was used as the threshold throughout the experiment. This value was obtained by observing a large amount of laser data. The signal strength needs to be sufficiently high to avoid unreliable measurements without rejecting good results. Then, a Kalman filter (KF) is applied to the readings from the laser range finder to further reduce the noises. Kalman filter is a recursive processing scheme that estimates the state of a dynamic system with noisy measurements, which are assumed to be zero-mean Gaussian white noise [34]. A process can be expressed with difference equations [34]: ( xk ¼ Axk1 þ Buk1 þ wk1 ð1Þ zk ¼ Hxk þvk where x 2 Rn is the state of a process and uk1 is the input while z 2 Rn the measurement. In our application, x is composed of the distance x1 and the rate of change in distance with respect to the change in rotation angle x2, i.e. x ¼ ½x1 x2 T ¼ ½x1 x_ 1 T . z is the reading of the laser range finder. Random variables wk and vk represent the process and measurement noise respectively, which are assumed to be independent, white, and normally distributed. The control input, Buk1 , in this application is equal to 0 because the pipe scanning robot travels along a straight line. When the laser sensor is on an actual robot system maneuvering through the pipe, the control inputs and measurements of veering off of the pipe axis could be used herein to increase the accuracy of the prediction model. Therefore, Buk1 ¼ 0 and A is expressed as   1 Dy A¼ ð2Þ 0 1 where Dy is the angle difference between measurements. The theory behind Kalman filter is well established and detailed information can be found in [34].

2.3. Center alignment The center of the pipe can be determined by three or four points distributed on the circle. Different from the tri-point method in [12], the intersection of multiple lines may give a better result as the center is derived from an average of multiple estimates. As illustrated in Fig. 1, four points ðxi ,yi Þ (i ¼ 1, . . . ,4) in coordinate XOY associated with two lines (l1 and l2) will intersect at the center of the circle O0 . The origin point O is the location of laser source. Let ðx0 ,y0 Þ and ðx00 ,y00 Þ are the center of the two lines l1 and l2 respectively. The circle center O0 is determined by the intersection of line l01 and l02 .

Fig. 1. The alignment of pipe center with laser measurements.

Z. Liu, D. Krys / Mechanical Systems and Signal Processing 31 (2012) 246–257

The coordinate of O0 ðx,yÞ, can be derived from the following equations: 8 ðx0 xÞðx2 x1 Þ > > þ y0 > > y ¼ ðx0 xÞðx4 x3 Þ þ y0 > : 0 y4 y3 The solution of Eq. (3) is 8 k1 x00 k2 x0 k1 k2 ðy0 y00 Þ > > > > > :y¼ k1 k2 where there are 8 y2 y1 > > < k1 ¼ x x 2 1 y4 y3 > > : k2 ¼ x x 4 3

249

ð3Þ

ð4Þ

ð5Þ

This method can be used with more than two lines. The center is calculated using each possible combinations of two lines and then the average is found between the resulting centers. Two points for each line must be chosen. The length of the line connecting these two points on the circle should be long enough so that noise effect on finding the center of the circle is minimized. 2.4. Process laser range image for anomaly detection The purpose of processing laser range image is to highlight artifacts or anomalies on pipe inner surface as potential problem locations. Two steps, i.e. laser image processing and segmentation, are involved in the procedure. The details are illustrated with flowchart in Fig. 2. The whole processing procedure consists of two steps: laser image processing and laser image segmentation, which are explained in detail in the following two subsections. 2.4.1. Laser image processing The first step is to run the data through a Gaussian low-pass filter and then ‘‘bad data’’ is removed. The ‘‘bad data’’ is identified by unrealistic values. Since the diameter of the pipe is known, any value that is much smaller or larger than this indicates an error in measurement.

Fig. 2. Procedure used in this paper to process laser range image.

250

Z. Liu, D. Krys / Mechanical Systems and Signal Processing 31 (2012) 246–257

Next is to up-sample the data and interpolate it using a least square method to generate a consistent grid of data [14]. The up-sampling helps smooth the laser range image. Once the data has been put into a uniformed grid, a median filter is applied to remove the salt and pepper noises in the image.

2.4.2. Laser image segmentation In the second step as illustrated in Fig. 2, the median filtered laser range image is first used to calculate its phase congruency map. Then, a morphological operation is applied to remove the small isolated regions. The rest of the regions and holes are filled. Finally, the segmentation operation is carried out to identify the regions of possible anomalies. To further characterize these regions, classification-based operation could be applied [21,9]. To characterize anomalies in the pipe profile, post-processing needs to extract geometric information from the pre-processed laser profile. Since the laser range finder measures the distance, segmentation with a threshold is a straightforward way to isolate the anomalies. However, thresholding with a fixed value does not work well for the processed laser range image due to the intensity inhomogeneity. Therefore, an image segmentation algorithm using local binary fitting (LBF) active contour model is employed to process the laser range image [35]. In our investigation, we found that the LBF active contour based segmentation alone did not work well for the laser range image in this application, given an arbitrary contour at initial stage. Therefore, we adopted a phase congruency (PC) algorithm to provide the initial contour as required by the image segmentation. The phase congruency is an absolute image feature measurement [36], which gives a value between 0 and 1 for each pixel in an image. One stands for the most salient feature while zero means no feature available at this point. A local energy model developed by Morrone and Owens [37] reveals that features are perceived at points in an image where the Fourier components are maximally in phase. A wide range of feature types give rise to points of high phase congruency. Kovesi proposed a scheme to calculate phase congruency using logarithmic Gabor wavelets [36,38]. The following equation is derived [36]: P P W o ðxÞbAno ðxÞDFno ðxÞT o c PCðxÞ ¼ o n P P ð6Þ o n Ano ðxÞ þ e

      

o: orientation; n: scale; E: small positive constant; Wo(x): the weighting function; Ano(x): the amplitude of Fourier series expansion; To: empirical compensation for noise; DFno ðxÞ: phase deviation

where bc denotes that the enclosed quantity is not permitted to be negative. The phase deviation DFðxÞ is defined as

DFðxÞ ¼ cosðfn ðxÞf ðxÞÞ9sinðfn ðxÞf ðxÞÞ9

ð7Þ

where fn ðxÞ and f ðxÞ are the phase angle and overall mean respectively. The PC for a two-dimensional image is calculated by summing the one-dimensional result over ‘‘o’’ orientations. A noise compensation To is performed in each orientation independently. For an extensive discussion of the underlying theory of phase congruency, Refs. [36,39] are recommended. With the contour derived from image phase congruency, the LBF active contour model is to minimize an energy function: Z eðC,f 1 ,f 2 Þ ¼ eLBF ð8Þ x ðC,f 1 ðxÞ,f 2 ðxÞÞ dx O

LBF x

where e is a local fitting energy function for a center point x 2 O. C is the contour in the image domain O. f 1 ðxÞ and f 2 ðxÞ are two numbers that fit image intensities near point x. The eLBF function can be expressed as x Z Z 2 2 eLBF KðxyÞ9IðyÞf 1 ðxÞ9 dy þ l2 KðxyÞ9IðyÞf 2 ðxÞ9 dy ð9Þ x ðC,f 1 ,f 2 Þ ¼ l1 inðCÞ

outðCÞ

where l1 and l2 are positive constants and K is a Gaussian kernel function. IðxÞ is a vector valued image. Eq. (9) indicates that when y approaches x, KðxyÞ returns a larger value, which accordingly has a greater impact on the difference of the intensities, i.e. 9IðxÞf 1;2 ðxÞ9. For each center point x, the local fitting energy eLBF can be minimized when contour C is x exactly on the object’s boundary [35]. In the experiment, the parameters l1 and l2 in Eq. (9) were set 1.0. The scale parameter s for the Gaussian kernel function was 3.0.

Z. Liu, D. Krys / Mechanical Systems and Signal Processing 31 (2012) 246–257

251

Fig. 3. Experimental setup. (a) Schematic of experimental setup. (b) Photograph of the experimental setup.

3. Experimental results 3.1. Experimental setup The experimental setup is shown in Fig. 3. A linear actuator is held up with two jacks. This allows the linear actuators height and angle to be adjusted. The cart of the linear actuator carries the laser range finder through the pipe. The laser range finder is connected to a PC-104 computer using a specialized high speed data acquisition card. A number of objects of varying size and shape are attached to the inner surface of the pipe to simulate surface abnormalities as shown in Fig. 4. The specification of each objects are listed in Table 2.2

3.2. Experiments for center alignment 3.2.1. Simulation results The simulation data was generated based on the setup of rpm (rotation per minute)¼1000 and sample rate¼20,000 sample/second. The diameter of the pipe was set as 82.55 cm. For the simulation, it is assumed that there are no bad readings, i.e. all data has an amplitude greater than 30, and the accuracy of the laser is 70:765 cm. Also 500 tests were preformed for each simulation to prevent any of the results skewed by the noise. It is also noted that the sensor is placed at a randomly generated location within 18 cm from the center of the circle. Table 3 shows the Kalman filter with different settings of noise and Ss, i.e. the standard deviation of the prediction model, which indicates the reliability of the model [34]. The error refers to the distance error (in mm) from the center of the circle. These results clearly show the benefits of using a Kalman filter to remove the noise from the range measurements. The value Ss ¼10.16 provided the best balance between the prediction model and the measurement. Fig. 5 illustrates the effect of using different numbers of lines to calculate the center of the circle. All tests were done using a Kalman filter with Ss ¼ 0.4. This plot shows diminishing returns for using more than 15 lines, which appears to be a good compromise for calculating the center of the circle.

2 There are three objects in category (4) and only the biggest is given in Table 2. Object (3) and (6) are not cuboid and the maximum dimension is provided in Table 2.

252

Z. Liu, D. Krys / Mechanical Systems and Signal Processing 31 (2012) 246–257

Fig. 4. Foam blocks used in the experiments. (a) The setup of foam blocks. (b) The geometry of foam blocks (in mm). Table 2 The specification of objects (in mm). Object

Length

Weight

Height

1 2 3n 4 5 6n 7

207 206 207 10 115 76 192

50 51 64 10 26 11 20

46 5 53 10 7 11 13

Table 3 Comparison of errors without and with a Kalman filter (unit mm). Noise added to distance

2.540 5.080 7.620 10.160

Error (without KF)

0.2540 1.0109 2.2809 4.0640

Error (with KF) Ss ¼ 5.08

10.16

15.24

20.32

0.0381 0.1448 0.3175 0.5410

0.0381 0.1194 0.2337 0.3759

0.0457 0.1321 0.2540 0.3988

0.0508 0.1499 0.2794 0.4394

3.2.2. Real test results Four experiments were preformed using real data taken from the laser range finder inside pipe (see Fig. 3). Each experiment was performed 3 times. The track was moving at a rate of 31.6 mm/s. Experiment 1 represents tests done in an empty pipe with the track centered. Experiment 2 is the same as Experiment 1 but with objects inside the pipe (see

Z. Liu, D. Krys / Mechanical Systems and Signal Processing 31 (2012) 246–257

253

Error vs Number of Lines

3.5

noise = 0.25 cm noise = 0.51 cm noise = 0.76 cm noise = 1.0 cm

3

Error (mm)

2.5 2 1.5 1 0.5 0

0

10

20 30 Number of Lines

40

50

Fig. 5. Error vs number of lines.

Cross Section of Pipe

400

400

300

300

200

200

100 0 −100 −200

100 0 −100 −200

−300

−300

−400

−400

−500 −500

0 distance [mm]

Cross Section of Pipe

500

distance [mm]

distance [mm]

500

500

−500 −500

0

500

distance [mm]

Fig. 6. Cross section of laser scan. (a) A smooth surface. (b) With objects on surface. Table 4 Test with real data for pipe center alignment. Setup

Test no.

Error (Unit: mm) Mean

Min

Max

Exp. 1

Empty pipe

1 2 3

2.293 2.296 2.352

1.843 2.024 1.761

2.755 2.621 3.051

Exp. 2

With objects in pipe

4 5 6

3.874 4.534 4.647

3.435 3.280 4.218

4.416 6.408 5.001

Exp. 3

Track in the empty pipe is 2.45 cm below the center

7 8 9

0.904 0.797 1.612

0.641 0.582 0.128

1.225 1.086 5.996

Exp. 4

Track in the empty pipe is 3 cm right to the center

10 11 12

1.801 1.554 1.521

1.487 1.011 0.752

2.929 2.068 2.568

Figs. 4(a) and 6). The purpose is to see how the presence of anomalies on pipe surface can affect the center alignment result. Experiment 3 is the same as Experiment 1 except the track is 2.45 cm below the center of the pipe. Experiment 4 is the same as Experiment 1 except the track is 3 cm right of the center of the pipe.

254

Z. Liu, D. Krys / Mechanical Systems and Signal Processing 31 (2012) 246–257

From Table 4, it is seen that surface anomalies do introduce more error to the center alignment (Experiment 2) in comparison with the experiments in empty pipe. However, compared to the size of the pipe, where d ¼ 82:55 cm, the maximum error, 0.5 cm, is just 1.2% of the radius. Shifting the laser source to the right and bottom produced a similar error to the one at the center and no big difference was observed. 3.3. Experiments for pipe profiling To test the profiling algorithm a number of varying size and shape foam blocks were taped to the inside of the pipe. Fig. 4 shows a few of the blocks used in the experiment. The original pipe profile acquired by the laser range finder and the 0

380 360 340 320 300 280 260 240

Distance (mm)

50 100 150 200 250 300 0

50

100

150

200

250

300

350

(mm)

Rotation angle (degree) 0

380 360 340 320 300 280 260 240

Distance (mm)

50 100 150 200 250 300 0

50

100

150 200 250 Rotation angle (degree)

300

350

(mm)

Fig. 7. The original pipe profile and the pre-processed result. (a) Original pipe profile. (b) Pre-processed pipe profile.

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0.5 0.4 0.3 0.2 0.1 0 −0.1 Fig. 8. The feature maps of pipe profile with phase congruency measurement. (a) The phase congruency map. (b) Processed PC map with morphologic operations.

Z. Liu, D. Krys / Mechanical Systems and Signal Processing 31 (2012) 246–257

255

Initial Contour

200 iterations

(3) (5)

(6)

(2) (1)

(7)

Fig. 9. Segmentation of laser scan with the method of LBF active contour model. (a) The initial contour derived from PC map. (b) Segmentation result after 200 iterations. (c) Labeled image of the segmented result.

pre-processed result are given in Fig. 7(a) and (b) respectively. The dark region on the left side is the bottom of pipe and is blind to the laser due to the support of the mirror. In the pre-processing, salt and pepper noises were removed. Fig. 8 presents the phase congruency map and processed one with morphologic operations ‘‘erode’’ and ‘‘fill’’. The regions of anomalies were roughly identified. The filled PC map was subtracted with a constant value to generate the initial contour map for segmentation. Herein, a constant value 0.15 was used. Fig. 9(a) and (b) gives the initial contour used for segment and the result after 200 iterations, respectively. The regions identified from the contours is shown in Fig. 9(c). Except objects in category number four, in which the maximum cube is of size 1 cm, all other six objects were successfully segmented. The category number four in Fig. 4 is composed of three small foam blocks. The bigger two can be observed from both the pre-processed result and corresponding phase congruency map. However, the segmentation algorithm missed them all due to their small size. Compared with object number two and five, the height of the biggest object in category number four is not the smallest. However, the area is small and was not captured by the segmentation algorithm. 4. Summary In this study, the feasibility of using a laser range finder on a robotic platform for vehicle body center alignment and pipe surface anomaly detection is investigated. The procedures to process laser range measurement data are developed. The segmentation of laser range image is implemented with image phase congruency measurement and region-based active contour model and the information about surface anomalies presented in the pipe profile is successfully extracted in the segmented image. To further understand the capability of the laser range finder, a study on the probability of detection needs to be carried out.

256

Z. Liu, D. Krys / Mechanical Systems and Signal Processing 31 (2012) 246–257

The use of foam objects in the experiments to simulate pipe anomalies reflects only one type, namely convex anomalies. In reality, a pipe is likely to have both convex (due to encrustation) and concave (due to loss of material). Nevertheless, this will not affect the laser scan and the processing algorithms need to be modified. The experimental results have demonstrated the effectiveness of the laser range finder for the inspection of de-watered pipes. The situation becomes more complex for an in-service water main. The attenuation to the laser beam due to the water becomes a problem. Special laser source for underwater use is expected. Besides, the reflectivity of the pipe inner surface will also have an impact on the quality of the laser range data. As described earlier in this paper, multiple NDI techniques are preferred for the pipe condition assessment. The laser profiling technique is a kind of visual inspection, with which subsurface discontinuities cannot be detected. Anomaly on the pipe surface may imply the final step to a failure. A second NDI technique is required to verify the sub-surface conditions. Even for the surface scanning, individual technique is constrained by its own limitation, the inspection could be further optimized by combining or fusing multiple complementary techniques so that the reliability and accuracy of the inspection can be improved. For future work, the inspection for in-service pipe is planned, where all the NDI sensors including the laser range finder need to work underwater. The factor of the absorbtion of laser light in water will be considered. Complementing other inspection techniques, like a machine vision system based on digital cameras, will be investigated as well. References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27]

[28] [29] [30] [31] [32] [33]

Y. Kleiner, B.B. Rajani, Comprehensive review of structural deterioration of water mains: statistical models, Urban Water 3 (3) (2001) 131–150. B.B. Rajani, Y. Kleiner, Comprehensive review of structural deterioration of water mains: physically based models, Urban Water 3 (3) (2001) 151–164. Smart Light Devices, Innovative subsea laser and imaging technology /http://www.sldltd.comS, retrieved in December 2011. Cool Vision, 3D profiling accuracy and reliability /http://www.coolvision.ca/S, retrieved in December 2011. CUES, CUES laser profiler /http://www.cuesinc.com/Laser-Profiler.htmlS, retrieved in December 2011. Optical Metrology Center, OMC laser profiler 15/50 /http://www.optical-metrology-centre.comS, retrieved in December 2011. Hancor, Laser profiling of flexible pipe /http://www.hancor.comS, Technical Note, 2007 (retrieved in December 2011). A. Dettmer, D. Hall, H. Hegab, M. Swanbom, Refining laser profiling methods used for pipeline assessment, in: North American Society for Trenchless Technology (NASTT) NO-DIG, Orlando, FL, 2005, pp. 1–9. O. Duran, K. Althoefer, L.D. Seneviratne, Pipe inspection using a laser-based transducer and automated analysis techniques, IEEE Trans. Mechatronics 8 (3) (2003) 401–409. K. Matsui, A. Yamashita, T. Kaneko, 3-d shape measurement of pipe by range finder constructed with omni-directional laser and omni-directional camera, in: IEEE International Conference on Robotics and Automation, 2010, pp. 2537–2542. G. Zhang, J. He, X. Li, 3D vision inspection for internal surface based on circle structured light, Sensors Actuators A: Phys. 122 (1) (2005) 68–75. M. Johnson, G. Sen Gupta, A robotic laser pipeline profiler, in: IEEE Instrumentation and Measurement Technology Conference, vol. 2, 2003, pp. 1488–1491. E. Wu, Y. Ke, B. Du, Noncontact laser inspection based on a PSD for the inner surface of minidiameter pipes, IEEE Trans. Instrum. Meas. 58 (7) (2009) 2169–2173. Z. Liu, D. Krys, B. Rajani, H. Najjaran, Processing laser range image for the investigation on the long-term performance of ductile iron pipe, Nondestr. Test. Eval. 23 (1) (2008) 65–75. P. Unnikrishnan, B. Thornton, T. Ura, Y. Nose, A conical laser light-sectioning method for navigation of autonomous underwater vehicles for internal inspection of pipelines, in: OCEANS 2009: Balancing Technology with Future Needs, Bremen, Germany, 2009, pp. 1–9. A. Nassiraei, Y. Kawamura, A. Ahrary, Y. Mikuriya, K. Ishii, A new approach to the sewer pipe inspection: fully autonomous mobile robot ‘‘KANTARO’’, in: 32nd Annual Conference on IEEE Industrial Electronics, 2006, pp. 4088–4093. A. Ahrary, A.A. Nassiraei, M. Ishikawa, A study of an autonomous mobile robot for a sewer inspection system, Artif. Life Robotics 11 (2007) 23–27. D.-H. Lee, H. Moon, H.R. Choi, Autonomous navigation of in-pipe working robot in unknown pipeline environment, in: IEEE International Conference on Robotics and Automation, 2011, pp. 1559–1564. D.-H. Lee, H. Moon, H.R. Choi, Landmark detection of in-pipe working robot using line-laser beam projection, in: International Conference on Control Automation and Systems, 2010, pp. 611–615. B. Zhuang, W. Zhang, D. Cui, Noncontact laser sensor for pipe inner wall inspection, Opt. Eng. 37 (5) (1998) 1643–1647. S.K. Sinha, F. Karray, Classification of underground pipe scanned images using feature extraction and neuro-fuzzy algorithm, IEEE Trans. Neural Networks 13 (2) (2002) 393–401. Z. Wang, Z. Hui, T. Wei, T. Yan, A new structured-laser-base system for measuring the 3D inner-contour of pipe figure components, Russ. J. Nondestr. Test. 43 (6) (2007) 414–422. J.-S. Yoon, M. Sagong, J. Lee, K. sung Lee, Feature extraction of a concrete tunnel liner from 3d laser scanning data, NDT & E Int. 42 (2) (2009) 97–105. M. Ritter, C.W. Frey, Rotating optical geometry sensor for inner pipe-surface reconstruction, in: D. Fofi, K.S. Niel (Eds.), Proceedings of SPIE: Image Processing: Machine Vision Applications III, vol. 7538, San Jose, CA, USA, 2010, pp. 1–9. W. Guo, L. Soibelman, J. Garrett, Jr., Automated defect detection for sewer pipeline inspection and condition assessment, Autom. Constr. 18 (5) (2009) 587–596. T. Yoshizawa, T. Wakayama, Development of an inner profile measurement instrument using a ring beam device, in: K. Harding, P.S. Huang, T. Yoshizawa (Eds.), Proceedings of SPIE: Optical Metrology and Inspection for Industrial Applications, vol. 7855, Beijing, China, 2010, pp. 1–8. A. Yamashita, K. Matsui, R. Kawanishi, T. Kaneko, T. Murakami, H. Omori, T. Nkamura, H. Asama, Self-localization and 3-d model construction of pipe by earthworm robot equipped with omni-directional rangefinder, in: IEEE International Conference on Robotics and Biomimetics, Phuket Island, Thailand, 2011, pp. 1–7. K. Matsui, A. Yamashita, T. Kaneko, 3-D shape reconstruction of pipe with omni-directional laser and omni-directional camera, in: Asian Symposium for Precision Engineering and Nanotechnology, Kitakyushu, Japan, 2009, pp. 1–5. R. Gooch, T. Clarke, T. Ellis, A semi-autonomous sewer surveillance and inspection vehicle, in: IEEE Intelligent Vehicles Symposium, 1996, pp. 64–69. O. Duran, K. Althoefer, L. Seneviratne, A sensor for pipe inspection: model, analysis and image extraction, in: International Conference on Image Processing, vol. 2, 2003, pp. III-597–III-600 (vol. 2). O. Duran, K. Althoefer, L.D. Seneviratne, Automated pipe defect detection and categorization using camera/laser-based profiler and artificial neural network, IEEE Trans. Autom. Sci. Eng. 4 (1) (2007) 118–126. S. Ogawa, H. Kuwano, Laser power control in a pipe inner wall inspection system with high-speed laser beam scanning, IEEJ Trans. Sensors Micromach. 131 (2011) 381–387. Schmitt Measurement Systems, Inc., AccuRange-4000 Laser Rangefinder Datasheet, Schmitt Measurement Systems, Inc., 2003.

Z. Liu, D. Krys / Mechanical Systems and Signal Processing 31 (2012) 246–257

257

[34] G. Welch, G. Bishop, An Introduction to the Kalman Filter, Technical Report, University of North Carolina, Chapel Hill, NC, July 2006. [35] C. Li, C.-Y. Kao, J.C. Gore, Z. Ding, Implicit active contours driven by local binary fitting energy, in: Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, 2007, pp. 1–7. [36] P. Kovesi, Image features from phase congruency, Videre: A J. Comput. Vis. Res. 1 (3) (1999) 1–26. [37] M.C. Morrone, R.A. Owens, Feature detection from local energy, Pattern Recognition Lett. 6 (1987) 303–313. [38] P. Kovesi, MATLAB and Octave Functions for Computer Vision and Image Processing, School of Computer Science & Software Engineering, The University of Western Australia, available from: /http://www.csse.uwa.edu.au/  pk/research/matlabfns/S, 2009. [39] P. Kovesi, Image Features from Phase Congruency, Technical Report, University of Western Australia, 1995.