Design And Research For The Road Detection System Based On Cluster Computing

In recent years, with the growing of the highway pavement testing requirements,the road detection has a higher demand. Pavement Detection system based on a singlecomputer can no longer meet existing needs, to a certain extent affected the process ofhighway construction in China. Therefore, the application of high-tech development ofhigh-performance road detection system has a very important significance.Multi-functional road detection system of this study is the application ofhigh-performance cluster computing technology, parallel processing by constructing thecluster system of pavement image data, can significantly improve the computationspeed on the road image data to provide faster, more reliable the road surface imageprocessing data, pavement management, conservation decision-making.This paper studies a multi-functional road detection system based on clustercomputing, which since the formation of a local area network multiple computersthrough a switch interconnect, parallel programming environment and thecorresponding configuration of the toolkit is installed on each computer, then use theload message passing parallel programming MPI integrated development environmentMicrosoft Visual C++6.0cluster system of the parallel program design, to build amulti-functional road surface detection system based on cluster computing hardwareenvironment and software platforms. Communication protocol using TCP/IP networkprotocols, communications applications, MPI between nodes in the cluster system toachieve, MPI provides non-blocking sending and complete communication function,which effectively reduces the communication overhead of the system, greatly reducingthe traditional stand-alone system processing time. The task allocation block allocationstrategy, the scheduling management node according to the task allocation strategy forthe division of tasks, and tasks are assigned to each compute node. Handlingcharacteristics, combined with the road surface image data is analyzed and summarizedthe findings of the existing load balancing algorithms, and design based on taskmigration, dynamic load balancing algorithm, the cluster system during operation to theinformation load of each computing node status based on reasonable computing tasksmigrate between the individual compute nodes only when N_i=0Properties N_j=0(where i!=j), the load balancing of the cluster system to start the re-load the heaviestnode p and the load node i lightest task allocation, load balancing at the same time tocomplete the calculation, T_i=T_p and the lightest load the compute nodes to receive theheaviest load of compute nodes migrate to the computing tasks, making the load of each computing node do possible equilibria.Finally, on a test cluster system has been developed to complete the road, thepavement images are collected by the “GuoChang” road detection vehicle. For example,computing tasks for different sizes, respectively, in a stand-alone system. There is nodynamic load balancing cluster systems and experiments to achieve dynamic loadbalancing cluster system, and according to the performance of the cluster system ofparallel speedup and efficiency evaluation of the computational performance of thecluster system.