Large-scale Fast Structured Light 3D Measurement System Based On Multi-view Geometric Constraints

Driven by the progress of science and technology,3D imaging technologies are playing increasingly important roles in both industrial and non-industrial fields.Among them,the 3D imaging technology based on structured light has been adopted by more and more 3D imaging systems due to its reliable measurement accuracy,convenient measurement method and stable measurement efficiency.However,in current three-dimensional imaging systems,the contradiction that measurement speed,measurement accuracy and measurement range cannot be taken into account at the same time is still the key factor that restricts the measurement effect.In addition,some objective factors,including high cost and technology blockade,also have an impact on the wide application of 3D imaging systems in all walks of life.To address above problems,we designed and developed a large-scale rapid structured-light3 D measurement system based on multi-view geometric constraints.The system consists of three major parts: hardware system,software system and core algorithms.The algorithm part includes corresponding solutions to the defects of current 3D imaging systems,mainly as follows:1.To address the problem of slow measurement speed,we innovatively introduce a composite coding-based stereo phase unwrapping method,which significantly improves image utilization,shortens the length of image sequence required for a single measurement,and reduces the single measurement time to 0.033 seconds.It is also supplemented with a deep learning optimization strategy to further improve the phase unwrapping accuracy.Therefore,the system can be applied to industrial assembly line to improve the efficiency of inspection.2.To address the problem of small measurement range,we use GPU accelerated multichannel parallel data-processing technology to reasonably plan the computation,greatly improve the speed of image processing,so as to achieve fast large-scale 3D measurement,and make sure the system can be widely used in the measurement of large objects.3.To address the problem of low measurement accuracy in motion scenes,the pixel-bypixel dynamic depth-constraint is used to exclude the error points while ensuring that the valid point cloud is not mistakenly deleted.In addition,by adding a motion compensation scheme,good stability of 45 micron is provided for the measurement of motion scenes.In the field of human-computer interaction,which requires high real-time performance,this technique can provide stable real-time 3D data.The hardware part makes reasonable planning for the layout of the device to achieve the best imaging effect,the software part optimizes the encapsulation of the algorithm,and makes good deployment of data processing and transmission to prevent data from duplication and repeated processing.In addition,the software also provides additional functions such as dynamic accuracy analysis,simple facial alignment,line drawing analysis and so on,which makes the application field of this system broader.In addition to completing the design and development of the above core content,this paper tests the core indicators of the system,and verifies the reliability of each function in actual scenes.