Vision-Based And 3D Laser Scanner Methods Of Pothole Detection

Roads play a very important role in the socio-economic development of the country. Road transport isquicker, more convenient and more flexible. A pothole is astructural failure in a road surface, usually asphalt pavement,due to water in the underlying soil structure and traffic passingover the affected area. Water first weakens the underlying soil;traffic then fatigues and breaks the poorly supported asphaltsurface in the affected area. Continued traffic action ejectsboth asphalt and the underlying soil material to create a holein the pavement. Detecting potholes is one of important tasksfor determining proper strategies of asphalt-surfaced pavementmaintenance and rehabilitation. However, manually detectingand evaluating methods are expensive and time-consuming.

Introduction

Roads are an integral part of the transport system. Acountrys road network should be efficient in order to maxi-mize economic and social benefits. Since roads have a greatinfluence on people’s lives, maintenance and management ofroads should be done periodically and exhaustively. However,due to lack of financial resources, many local governmentsare not able to conduct sufficient inspections Potholes mayresult from four main causes:

• Insufficient pavement thickness to support traffic dur-ing freeze/thaw periods without localized failures
• Insufficient drainage
• Failures at utility trenches and castings (manhole anddrain casings)
• Pavement defects and cracks left unmaintained andunsealed so as to admit moisture and compromise thestructural integrity of the pavement.

Potholes can grow to several feet in width, though theyusually only develop to depths of a few inches. If theybecome large enough, damage to tires, wheels, and vehiclesuspensions is liable to occur. Serious road accidents canoccur as a direct result, especially on those roads wherevehicle speeds are greater. The existing pothole detectionmethods are divided into three: Vibration-Based detectionmethods, 3D reconstruction-based methods and vision-basedmethods. Taking the graph of road accidents occured in Indiafrom the year of 2000 to 2013 are shown in Fig1. As one type of pavement distresses, a pothole is defined as a bowl-shaped depression in the pavement surface and minimum. Potholes can generate damagessuch as flat tire and wheel damage, impact and damage on thelower part of a vehicle, sudden braking and steering wheeloperation, and vehicle collision and major accidents.

Vibration-based pothole detection

This method use recent data acquisition hardware todevelop a vibration-based system for preliminary evaluationof pavement conditions. The pavement distress such as cracksand rutting impose impacting forces on the vehicle.Thepavement surface conditions can be estimated from therecorded responses of the testing vehicle when driving onthe pavement. The advantage of this detection system isof small storage requirement, cost-effective and amenablefor manageable real-time data processing. An extensiveroad network Sri Lanka that spans the country and newroads are being built every day, yet even the roads arenot maintained properly in the capital city. Vibration-basedmethod use accelerometers in order to detect potholes. Avibration based system has the advantages of these methods require small storage and can be used in real-time processing.

However, vibration based methods could provide wrong results that the hinges and joints of road can be detectedas potholes. Potholes in the center of a lane cannot be detected using accelerometers due to no hit by any of the vehicles wheels. Vibration-based methodsfor detecting potholes can be summarized as shown in fig 3.

3D reconsturction

The classification of 3D reconstruction methods can be further into three methods that are 3D laser scanner meth-ods, stereo vision methods, visualization using MicrosoftKinect sensor. A. 3D laser scanner methodsThe main idea behind the technique used by 3D laser scanner is that employs reflected laser pulses to create accurate digital models of existing objects. The accurate 3D point cloud points with their elevationswere captured during scanning and extracted focusing onspecific distress features by means of a grid-based processing approach.

In the study by Chang et al. For 3D laser points the segmentation method is dedicated to the features forroad pavements such as to automate the potholes identifi-cation. For obtaining a regular grid for this purpose TIN-based interpolation was employed. Although data quality would lose somehow after the process, regular grid was deemed appropriate and adopted for subsequent processing. For reducing noises Mean filter was applied to the grid. Afterwards, to extract the target features such as potholes, topographic, hydraulic, and texture features are applied. Finally, to integrate the pixels of similar features into regions,and thus, to obtain the area sizes of each potholes, clusteringmethod is used. Real-time, low-cost inspection system to detect distress features such as rutting, shoving and potholes using high-speed 3D transverse scanning techniques consisting of an infraredlaser line projector and a digital camera is introduced by Li etal. To improve the accuracy of the system, a multi-viewcoplanar scheme is employed in the calibration procedure sothat more feature points can be used and distributed acrossthe field of view of the camera.

Potholes can be detected in real time using Laser scanning systems. However, the costof laser scanning equipment is still significant at vehicle-leveland currently these works are focused on the local accuracy of 3D measurement.

Stereo Vision Methods

Two digital cameras are used to cover a pavement surfacein this method. To analyze 2D images from each of the two cameras to detect and classify any cracks is the first step. The results from analyzing two image sources of the same pavement are then combined so that cracks missed by one analysis are still counted, therefore potentially achieving higher accuracy. Also, the pair of images on the same pavement surface is used to establish 3D surface model with longitudinal and transverse profiles through geometric modeling. The sequence of steps such as camera calibration, distortion correct, matching stereo points, 3D reconstruct, and profile report should be performed, to recover the 3Dproperties from given pairs of 2D images on the same pave-ment surface.

Hou et al. presented a method of applying thestereovision technique into pavement imaging to reconstructthe 3D pavement surface from a pair of images. A totalof four cameras were used in two pairs to collect pavementsurface images across a 4-meter wide pavement (each pair ofimages covers 2 meters of the road). In this method, 4 stepssuch as calibration, distortion adjustment, matching and 3Dreconstruction was involved.

As field experiments, the road section was measured on alocal road with the length of 650m. Stereo vision methods need a high computational effort to reconstruct pavement surface through the procedure of matching feature points between two views so that it is difficult to use them in areal time environment. Also, both cameras should be very accurately aligned since the cameras may misalign and affect the quality of the outcome if there is the vibration by the vehicle motion.

Stereo Vision in Digital Highway Data Vehicle (DHDV)

The stereo vision techniques for the measurementof pavement condition with a stereo vision system attachedto a vehicle for recording of the road network conditionssuggested by Staniek. 3D reconstruction methods fordetecting potholes are shown in fig 4.

Kinect Sensor

A low-cost sensor system using Kinect sensor and ahigh-speed USB camera to detect and analyze potholesproposed by Joubert et al. Recently, a low-cost Kinectsensor to collect the pavement depth images and calculate 3D reconstruction methods for detecting potholes the approximate volume of a pothole discussed by Moaz-zam et al. The pavement depth images were collected from concrete and asphalt roads Using a low-cost Kinectsensor. For better visualization of potholes Meshes were generated. Area of pothole was analyzed with respect to depth. Using trapezoidal rule on area-depth curves throughpavement image analysis, the approximate volume of potholewas calculated. Although it is cost effective as compared toindustrial cameras and lasers, the use of infrared technologybased on Kinect sensor for measurement is still a novel ideaand further research is necessary for improvement in errorrate.

Vision-based pothole detection

2D Image-Based Approaches

A method for automated pothole detection in asphalt pavement images proposed by Koch and Brilakis. Underthis method, the image is first segmented into defect and non-defect regions. According to the geometric characteristicsof a defect region, the potential pothole shape is thenapproximated. Next, with the texture of the surrounding non-defect region, the texture of a potential region is extractedand compared. The region of interest is assumed to bepothole, if the texture of the defect region is coarser andgrainier than the one of the surrounding surface. It was implemented in MATLAB utilizing the Image ProcessingToolbox for the testing purpose. Using a remote-controlledrobot vehicle prototype equipped with a HP Elite AutofocusWebcam images were cropped from video files capturedwhich was installed at an altitude of about 2 feet.

A new unsupervised vision-based method which does not require expensive equipment, additional filtering and training phase proposed by Buza et al. Here uses image pro-cessing and spectral clustering for identification and roughestimation of potholes.

Video-Based Approaches

It cannot determine the magnitude of potholes for assess-ment since 2D image-based approaches have been focusedon only pothole detection and is limited to single frame. Torecognize a pothole and calculate total number of potholesover a sequence of frames Video-based approaches wereproposed. A method of 3D sparse reconstruction discussedby Golparvar-Fard et al., and 3D dense reconstructionand mesh modeling was based on the dense 3D point cloudreconstruction of Golparvar-Fard et al. The method wasvalidated on several actual potholes using a Canon Vixia HDcamera .method for automated detection and assessment of potholes,cracks and patches from video clips of Indian highwayspresented by Lokeshwor et al. Here first captured videoclips are segmented automatically into two different types offrames category. The frames with distress and frames withoutdistress using DFS algorithm. Then, database of frameswith distress is processed with CDDMC algorithm. Kochet al.presented an enhanced pothole-recognition method inorder to complement and improve the previous method,which updates the texture signature for intact pavementregions and utilize vision tracking to track detected potholesover a sequence of frames. The texture-comparisonperformance was increased by 53%, and the computationtime was reduced by 57% compared with the previousmethod. Here assumed that only one pothole enters theviewport at a time, and therefore additional work is neededfor considering multiple potholes in the viewport. Vision-based methods methods for detecting potholes are shown infig 5.Fig. 5. Vision-based methods for detecting potholes.

Conclusions

Detecting potholes accurately is one of crucial tasks fordetermining proper strategies of pavement maintenance. Butthe person manually detecting and evaluating methods arecostly and time-consuming. Vibration-based methods, 3Dreconstruction-based methods, and vision-based methods arethe main division of the existing pothole detection methods.

So from the study, the vision-based methods are cost-effective compared to 3D laser scanner methods, it may be difficult to accurately detect a pothole by the discussed methods. Due to the misrepresent signal generated by noisesince they detect a pothole through analysis of the collected image and video data. Therefore, it is need to develop a pothole detection method using different features in 2D images which improve the present pothole detection method and can accurately detect a pothole.