三維量測廣泛應用於工業檢測、智慧製造與自動化量測，但中遠距離條件下的投影能量、影像取樣與系統幾何仍可能限制量測品質。本研究提出以投射式雷射散斑特徵位移進行中遠距離三維物體重建與辨識的方法，並以結構光掃描點雲作為幾何參考模型。 本研究以 532 nm 綠光雷射經物鏡擴束後照射擴散片形成散斑圖樣，利用二維自相關、徑向平均與半高位置量測散斑尺寸（Speckle Size），分析傳播距離與有效照明孔徑的影響。其後以 0–30 cm 多平面校正與正規化互相關（Normalized Cross-Correlation， NCC）追蹤散斑特徵，將二維位移軌跡投影至主要方向並以一階模型建立深度對應；未知物重建時，於搜尋區域（Search Area）內選取最高 NCC 位置估測高度並建立三維點雲。 重建點雲三軸統一為毫米後，以 K = 16 的 K 近鄰（K-Nearest Neighbors， KNN）平均鄰近距離與中位數絕對偏差（Median Absolute Deviation， MAD）門檻移除離群點。辨識流程以快速點特徵直方圖（Fast Point Feature Histograms， FPFH）建立候選對應，經隨機抽樣一致性（RANSAC）粗配準與點對面迭代最近點（Point-to-Plane Iterative Closest Point， ICP）精配準，再以均方根誤差（Root Mean Square Error， RMSE）評估幾何差異。 實驗結果顯示，散斑尺寸隨傳播距離增加而增大，並隨有效照明孔徑增加而減小；在 6 m、12 m 與 18 m 的測試配置下，六張面具皆能以最低 RMSE 正確匹配至對應參考模型。固定 12 m 時，測試範圍內較小的散斑尺寸整體上對應較低的正確配對 RMSE，顯示本方法具有應用於中遠距離三維重建與已知模型辨識的可行性。;Three-dimensional (3D) measurement is widely used in industrial inspection， intelligent manufacturing， and automated metrology， while projected optical power， image sampling， and system geometry can limit measurement quality at mid-to-long ranges. This study proposes a method for mid-to-long-range 3D object reconstruction and identification based on projected laser speckle feature displacement， with structured-light point clouds used as geometric reference models. A 532-nm green laser is expanded by a microscope objective and projected onto a diffuser to generate a speckle pattern. Speckle size is measured from the two-dimensional autocorrelation， radial average， and half-maximum position to investigate the effects of propagation distance and effective illumination aperture. Multi-plane calibration images acquired over 0–30 cm are then used to track speckle features by normalized cross-correlation (NCC). The two-dimensional displacement trajectory is projected onto its principal direction and modeled by a first-order depth relation. During reconstruction， the highest NCC within a search area surrounding the calibrated trajectory is used to estimate the feature height and construct a 3D point cloud. The reconstructed coordinates are converted to a common millimeter scale. Outliers are removed using the mean three-dimensional distance to K = 16 nearest neighbors with a threshold defined as the median plus one median absolute deviation (MAD). Fast Point Feature Histograms (FPFH) establish candidate correspondences， Random Sample Consensus (RANSAC) provides coarse registration， and Point-to-Plane Iterative Closest Point (ICP) performs fine registration. Geometric discrepancy is evaluated by the one-way nearest-neighbor root mean square error (RMSE) from the speckle-reconstructed point cloud to each structured-light reference point cloud. Experimental results show that speckle size increases with propagation distance and decreases as the effective illumination aperture increases. Under the tested 6 m， 12 m， and 18 m system configurations， all six masks are correctly matched to their corresponding reference models by selecting the lowest RMSE. At a fixed distance of 12 m， smaller speckle sizes within the tested range generally correspond to lower correct-match RMSE values. These results demonstrate the feasibility of projected speckle， multi-plane calibration， and point-cloud registration for mid-to-long-range 3D reconstruction and closed-set object identification.
