The necessity for frequent underwater cable/pipe inspection is becoming more apparent with increased construction of subsea piping networks for the oil and gas industry and heavy international telecommunication traffic. Current methods for the surveillance, inspection and repair of undersea cables/pipes utilize remotely operated vehicles (ROV) controlled from the surface. This work can prove to be very tedious and time consuming while also being prone to human error due to loss of concentration and fatigue. Cables can be covered along sections of their length thus making it difficult to recover cable trajectory after losing track. A reliable image processing based cable tracking system would prove much less expensive and less prone to error than current solutions, as the need for constant operator supervision is removed. The development of the vision based cable tracking system for use on autonomous vehicles would also be beneficial because of the reduced cost as a mother ship is no longer necessary and such systems are beginning to appear in commercial use (Hydro-International 2008). Vision systems also possess advantages over magnetometer and sonar based solutions for cable tracking (Ito et al. 1994). Vision systems prove less expensive, have the ability to identify faults and require a smaller less powerful vehicle for operation (Ortiz et al. 2002).
An early attempt at a cable tracking system using machine vision was developed by Matsumoto and Ito (Matsumoto & Ito 1995). The method, like most underwater cable tracking techniques, takes advantage of the lack of straight line edges found in the
Fig. 2. Camera and lights setup (red box illustrates image frame)
underwater environment. An edge image of the sea floor is acquired using a Laplacian of Gaussian filter. The Hough transform is then applied to the edge pixel image in order to find the most likely pipe edge candidates. A method of candidate evaluation is implemented by examining the length and width of each edge pixel line candidate. The direction of the cable in the present image and the previous image are used to predict the angle of the Hough transform to be applied to the subsequent image to reduce computation time. This cable following algorithm also attempts to address the problems of sediment covered pipes and non-uniform illumination. While achieving reasonable results in a controlled environment, factors such as spurious edge detection from other pipes or elements, abrupt pipe direction changes and a search algorithm (when cable is undetected) have not been accounted for and result in reduced performance.
Balisuriya et al. developed on previous work (Balasuriya et al. 1997) by adding an a priori map of the cable location to his technique (Balasuriya & Ura 2002). The main features of the method are the ability to follow the cable when it is not visible to the vision system and selection of the correct cable in the image (in the case of multiple cable presence). These objectives are addressed by assuming that an a priori map of the cable is available. The a priori map serves three purposes; to predict the region of interest (ROI), to avoid misinterpretations with other cables in the image and to be used as a navigation map in the case where the cable disappears from view. A similar method to Matsumoto and Ito is implemented to locate the cable in the image by utilizing the Hough transform. The technique described fuses inputs other than optical information to track the cable and has attempted to overcome the issues of tracking a cable when it becomes partially or fully obscured to the vision system (due to sediment or algae coverage). It also addresses the difficulty associated with correct cable selection. The method demonstrates that the extra information, in the form of a map, fused with optical sensing can greatly improve performance. Unfortunately, having an a priori map of the cable location is not always a realistic assumption especially in the case of older installations.
Ortiz et al. developed a method for real-time cable tracking using only visual information that again takes advantage of the cables shape to locate strong alignment features along its side (Ortiz et al. 2002). After the initial image segmentation step the contour pixels are examined to locate pixel alignments that display strong pipe characteristics (long pixel alignments, parallel alignments and alignments in a y direction on the image). Once the cable has been located in the image a Kalman filter is implemented to reduce the ROI for the subsequent image to reduce computation time. When anomalies occur in the prediction phase actions are taken in order to correct the algorithm; either the frame causing the anomaly is discarded or, if a number of consecutive frames are incorrect, the Kalman filter is reset. This method achieved a 90 percent success rate for trials at 25 frames/sec performed on old cable installations. The technique dealt reasonably well with partially covered cables however, a minimal presence of the cable is required in the image at all times. No backup system in the scenario where the cable becomes undetectable by the system is described. The performance of the method discussed by Ortiz et al. (Ortiz et al. 2002) was later improved upon while also reducing the complexity of the system (Antich & Ortiz 2005). This new technique also includes a first approximation to the vehicle control architecture for locating and tracking the cable autonomously using the vision system and a method is proposed for unsupervised tuning of the control system. Both the control system and the tuning strategy were validated using 3D object-oriented simulator implemented in C++ using the OpenGL
graphics library. Only simulation results have been published to date but results for the implemented control architecture are promising.
Recently Wirth et al. developed a method for cable tracking by implementing a particle filter in an attempt to predict the location of the cable when it is partially obscured and thus the number of extracted image features is reduced (Wirth et al. 2008). A motion model is calculated to describe the cable parameters' changes over time using previously captured cable inspection footage. An observation model is also described to detect cable edges in the image. These models are then combined in a particle filter which sequentially estimates the likelihood of the cable position in subsequent frames. Experimental results concluded that the system was capable of working online in real time and showed good performance even in situations where the cable was scarcely visible. A method for dealing with multiple cable presence has yet to be developed for the system.
Different methods for cable tracking systems exist each with their own advantages and disadvantages. The work reviewed uses similar techniques for cable detection (looking for straight line edges) but differ in their approaches to cable direction prediction to save on computational expense and improve detection robustness. The need for a robust system for tracking a cable that is partially obscured for a short segment remains a priority. Sensor fusion has been proved to be a good approach to robust cable following when the cable is in view. The future will focus on refining tracking methods and working towards the development of vision systems for inspection, fault identification and localisation with the hope of fully automating the process of cable tracking and inspection and reducing human input. There remains a lot of room for improvement in these systems but despite this there is a surprising lack of publications in the field over recent years.
Was this article helpful?