Mapping Underground Utilities with Complex Spatial Configuration Using Ground Penetrating Radar

Records of the locations, measures, and properties of over 20 million mileage underground utility networks in the US are incomplete, inaccurate, and many times unavailable. This lack of information of underground pipes and cables is a primary reason for over six million of utility interruptions every year that cause injuries, fatalities, property damages, and environmental pollutions, amounting to over eight billions of dollars in loss. It also poses a critical challenge to maintaining and upgrading the ageing underground infrastructure with a D+ score according to the American Society of Civil Engineers’ Report Card. There is a critical need for accurately mapping and labelling underground utilities. Prominent among methods in mapping underground pipes, ground penetrating radar (GPR) has proven its merits in quantitatively detecting, locating both metallic and nonmetallic buried utilities. GPR works as follows. It sends electromagnetic (EM) pulses via a transmit antenna to the underground and collects part of the reflected EM wave via the receiver antenna. The control unit measures the EM wave properties, which will be altered when it encounters changes in subsurface layers and buried anomalies. This process produces a two-dimensional data matrix with columns containing the reflected amplitudes at certain times for each measurement position along each trace. Since the radargram shows the patterns related to the location of underground utilities, an interpretation process is needed to restore spatial information from patterns. Various signal-processing or pattern recognition algorithms have been created to automatically or semi-automatically interpret the reflected signals or patterns from GPR radargrams. However, it is still constraint to apply current approaches to the utility-congested urban environment because of the complex spatial configurations of underground utilities and its interfered signal reflection in the resulting GPR scan, which are very difficult to interpret. Two main scenarios we are usually facing are: (1) the oblique pipe orientation produces irregular GPR patterns, and (2) adjacent pipes generate coupled or occluded GPR patterns. There is a research need to design a novel approach to automate the process of mapping underground utilities with complex spatial configurations from multiple transformed signatures using GPR system. The goal of this research is to create a novel approach to accurately and automatically detect and estimate underground utilities with complex spatial configurations. After reviewing the exiting studies and applications, I identified four knowledge gaps that have to be filled to achieve this goal. The first knowledge gap is the lack of a practical algorithm to automatically extract hyperbolas and segment them into legs, and peaks, which is the essential step for analyzing the patterns of GPR signatures from the scanned images. For instance, the peak of the hyperbola indicates the closest point to the GPR survey trajectory. The intersecting point between the right trailing leg of one hyperbola-shape GPR signature and the left rising leg of the other hyperbola-shape GPR signature indicates there may exist two pipelines close to each other. By analyzing the decomposed segments of the hyperbola shapes, the possible spatial configuration of the buried pipes can be estimated, providing an “educated” guess for the spatial configuration, size, and location of underground utilities in congested urban areas. The second knowledge gap is that the causality between the complex spatial configuration of underground utilities and resulting patterns in GPR scanned images has not been thoroughly established, which makes the inverse estimation from transformed GPR signatures to complex spatial configurations of buried utilities impossible. Therefore, in order to interpret the GPR data and map the underground utilities in an automatic manner, it’s essential to clearly understand the rationale that: (1) how the spatial configurations affect the transformation and occlusion of generating GPR signatures, and (2) how we can inversely estimate the spatial configurations from the transformed, occluded GPR signatures. The third knowledge gap is the lack of an intelligent GPR survey trajectory planning approach, in which GPR data are processed and interpreted in real-time and the trajectory is automatically adjusted correspondingly. The rationale is as follows. Ideally, the perpendicular-to-pipe scanning yields highest detectability, and along-pipe scanning yields highest planimetric and depth accuracy. However, it is quite challenging for field surveyors to maintain the ideal angels, i.e., “perpendicular-to-pipe” and “along-pipe”, in the survey grid while not knowing the exact orientation of the pipes. The deviation between the orientations of the buried pipes and the directions of the GPR moving trajectories will pose ill-shaped or incomplete signatures in GPR scanned images, which brings great challenges in the succeeding signal/images processing and utilities attribute estimation from the collected field data. If we can adjust the trajectory timely when we observe the ill-shaped or incomplete signature in the filed survey, we could guarantee a good angle and a reasonable accuracy. The fourth knowledge gap is the lack of an illusion-free visualization platform for information sharing and communication of buried infrastructure. Current AR platform is limited to visualize multiform information of buried utilities retrieved from GPR data in three aspects, i.e., (1) the unstable dynamic tracking of pipes in markerless environment causes visual fatigue; (2) the missing depth cues (e.g., relative size, occlusion, shadows) affect the quality of visual integration of the physical objects and virtual pipes underneath; and (3) the indirect way to differentiate materials of pipes in either color-labeled or rendering mode. With these limitations, the data sharing and communication efficiency are still underestimated in the current visualization platform. To overcome these knowledge gaps and arrive at the research goal, four research aims have been formulated. Aim 1: The first aim is to develop an algorithm to automate the detection and decomposition of GPR signatures into feature components, i.e., hyperbola apex, rising legs, trailing legs and junction points of intersecting hyperbolas. Accomplishing this aim will provide all the essential information needed for determining the complex spatial configurations and individual attributes (size, dimension, orientation, etc.) of buried utilities in a congested area. Besides, by utilizing the delivery in this aim, many existing algorithms can be enabled to automate underground utilities detection and mapping in utility-congested urban environments. Specifically, the so called “drop-flow” algorithm commences at a strip of pixels from the top of the edge of the scanned image, which mimics the motion of a “raindrop” falling or flowing as it touches the edge pixels of the image. The movement of the “raindrop” completes the decomposition of the GPR signature when it touches the “ground”, i.e., the bottom of the edge image. The algorithm was tested using both synthetic and field data, which generated a detection rate of 84% and a precision of 78%. The results show that this drop-flow algorithm is capable of differentiating hyperbolas and identifying the feature points and segments of each hyperbola. The algorithm has at least two outstanding merits: (1) there is no need for an initial guess of the number of hyperbolas, and (2) it is capable of not only detecting the number of hyperbolas, but also decomposing individual hyperbolas into rising leg, apex, and trailing leg, as well as the intersections between neighboring connected hyperbolas. Chapter Two of this dissertation is devoted to solving this problem. Aim 2: The second aim is to establish the causality between the complex spatial configurations of ..