School of Computer Science and Software Engineering Monash University Research Proposal — Semester 2 , 2005 [ Detection and Measurement of Pavement Cracking

The state of Victoria in Australia has many major roads assets. These roads, stretching about 196,000 in km, have to be maintained and managed by VicRoads, the road authority of Victoria[1], and the local councils. Pavement Management Services (PMS) [2] company was employed by VicRoads in 2002/3 to do the analysis and detection of the pavement cracking in Victoria. PMS uses up to five industrial quality digital cameras which are mounted in different directions on a host vehicle (Figure 1). Figure 1. PureData mobile digital video system. The resulting video has the resolution of 768 x 576 pixel, which is a typical broadcast quality[2]. Out of five cameras, one is used to obtain the detailed view of the pavement surface. This detailed views (Figure 2) are the images that contain the cracking and defect information on the roads. In this project, road cracks are defined as " a series of connected dark spots appearing against other marks in the background " [3] with the minimum width of 3mm. The original method of the crack detection employed by PMS is by traveling in a vehicle at 20 km/h. The driver watches the road for the cracks, and when a crack is found, another passenger of the vehicle records the crack. This original method is slow and unsafe [4]. PMS then employs a better method in crack detection, which is by recording images of the pavement at speeds up to 100 km/h [4]. The collected images are then inspected off-line by multiple pavement rating operators [4] to identify cracks on the footage. When a particular frame contains cracks, the operators mark the location of the frame. Although this present method has the advantage of reducing the time to do the assessment of crack detection, the manual method is still expensive, tedious, time consuming and can be inaccurate [4]. This project aims to design an algorithm able to automatically detect and locate road cracks in the road pavement video footage of Victoria. Initial work has been carried out by a previous honours student. This student proposed a modified Watershed algorithm to solve the problem of detecting road cracks [5]. The main challenge in detecting road-cracks in this project is the low-resolution of the video footage which creates noises in the video frames. The second challenge is the possibility of false detection of cracks. Shadow from trees, bridges, and tunnels pose significant problems to …