Double-laser-stripes Measurement System For Subway Tunnel Profile Based On High-speed Vision

Structure-light measurement with the merits of large range, large field of view, high precision, simple extraction method, real-time characteristic, has been widely used in traffic for three-dimensional information acquisition. Currently, this method is widely used on the inspection cars, to obtain the whole profile information of rail infrastructure. However, there arc many defects in the traditional single structure-light system. Thus, this paper presents a double structure-light scanning system. And this system is applied to the detection of the tunnel profile to get efficient and accurate measurement results.The measurement principle of the system, the work process and the selection of desired devices, the overall design of the system are introduced. Then, the main content are presented in the different chapter, including improved image processing algorithm and its FPGA implementation, PC program design and new measurement calibration method based on total station.On the basis of the analysis of common image processing algorithms, improved laser stripe extraction algorithm is stated, which can be used to process pictures at300fps, with a resolution of1280x1024, to realize real-time image processing and storage. The pipeline structure is used in this part, and the algorithm is transferred to the FPGA based on VHDL programming language. The extraction speed and accuracy is improved a lot, due to these methods.The difficulty of programming host computer is to achieve stable storage of data which is the process result of FPGA processing in real-time. The original system has half data to the existing system, and there are serious dropped frames. So multithreading is used in existing systems to store data fast and stable. From the analysis of experimental verification results, this is better than single-threaded.The acquisition of calibration point of system has always been a difficulty. Based on the actual tunnel detection situation, the method of using total station to obtain calibration point is proposed, which makes the acquisition of calibration point more efficient and effective. Then the BP neural network is utilized in fitting operation.From the analyzing of test results, this system has an average error less than3mm, in the range of about2m and functions well with little dropped frames, which can achieve the desired requirements.