Research On Methods And Evaluation Of Stripe Center Extraction In Structured Light3D Measurement

Structured light3D vision measurement, because of its non-contact, high precision, real-time, high controllability, wide measurement range, is widely used in the actual measurement. The structured light center extraction is a key step of the3D measurement. When the measurement system is fixed, structured light3D vision measurement error mainly depends on the light bar center extraction error. For actual measurements, especially in the outdoor large-scale measurement, extraction of structured light center is not only subject to random noise, but also by non-random noise such as clutter noise or structured noise. The reliability and accuracy evaluation of stripe center extraction is a problem to the actual measurement.On the basis of the structured light generating mechanism, this dissertation studied the structure and the principle of the line laser, illumination model and image sensors. According to the analysis of the structured light energy transfer process, the grey of the final structured stripe image is proportional to the value of the structure light energy. Different types of structured light samples are observed, including the stripe samples impacted by the objects’ shape, colors, materials, the reflected light, the camera defocus blur and background motion blur. The main factors are the strong absorption caused by color, material due to the strong diffuse and specular reflection interference. The different grey types of stripe cross-sections, fit the Gaussian model.The various existing stripe center extraction algorithms are based on the mathematical characteristics of the stripe cross-section. These algorithms have both advantages and disadvantages. A non-linear extraction algorithm is proposed in this dissertation, which has non-random interference robustness, and reserves the reliability evaluation of stripe center extraction.For the non-random noise disturbances, the reliability evaluation of stripe recognition is proposed. This reliability evaluation is based on the energy of stripe cross-section normalized to Gaussian model. Experiments show that this reliability evaluation’s results of the extracted stripe center can distinguish a variety of interference, and the evaluation results fit the actual situation very well. At the same time, this reliability evaluation system can be used to evaluate how reliable the stripe center extraction algorithms are to non-random noise disturbances. Among the algorithms, the algorithm proposed by this dissertation is best to resist non-random noise.For the impact of random noise, the precision evaluation for light stripe extraction is proposed. This dissertation theoretically derived the error variance of the stripe center position, and on basis of this, estimated the error statistical distribution. According to this distribution, the stripe center precision evaluation can be done, and by the way, the confidence evaluation based on the precision can be done too. The variance of the extracted stripe centers is inversely proportional to the image signal to noise ratio, and is related to the stripe width and filter kernel size. Experiments show the precision evaluation results are consistent with the actual situation. This dissertation evaluated the precision of stripe center extraction algorithms, and the algorithm proposed by this dissertation is better.Finally, this dissertation described the relationship between the precision evaluation and reliability evaluation of the extracted stripe centers, provided a reference for the accuracy and reliability of the actual measurement results, and also offered the basis for selection of the stripe center extraction algorithms.