Effective segmentation method of target based on polarization-modulated 3D imaging system

Abstract

To implement Real-time segmentation for polarization-modulated 3D imaging system, an efficient segmentation method for multi-dimensional information fusion is proposed in this paper. Conventional 3D imaging Lidar systems use an avalanche photodiode (APD) array detector to measure time-of-flight of each pixel in the scene. Here we propose and experimentally demonstrate a super-resolution 3D imaging framework based on a new imaging sensor EMCCD (Electron Multiplying Charge Coupled Device). Due to its low bandwidth characteristics, the electro-optic modulators are applied to implement temporal (range) resolution, and meanwhile act as a fast shutter with sub-nanosecond-level. Consequently, range-gated 3D imaging can be achieved to improve the signal-to-noise ratio (SNR) performance in our framework. With dual EMCCDs structure, the depth map and intensity image can be reconstructed from adding the two modulated images. The iterative threshold algorithm method is applied to the target segmentation of high-resolution images, and image morphological erosion algorithm are used to improve the segmentation accuracy. The target’s pixel coordinate position obtained by image segmentation is mapped to 3D point cloud data to get the segmented target point cloud data. Experimental results show that the system can achieve high-precision flash imaging. Meanwhile, the segmentation method has a great improvement in time efficiency compared with traditional clustering algorithm, and can reduce the under-segmentation error rate. Ultimately, we found that the imaging method showed outstanding performance on high-precision imaging with an error less than 0.1m in a wide field-of-view of 0.9mrad. And the segmentation of target takes only 560ms.