利用位能場做部分表面之三維點雲車體形狀辨識

Abstract

本論文之研究主要是利用位能場模型來做為三維汽車車體的形狀比對與辨識，此模型的做法是先將待測物縮小置於模型內部，再藉由待測物與模型彼此之間的推斥力與力矩作用位置調整及方向旋轉，再搭配放大及碰撞測試限制待測物必需在模型內，若兩者為相似的形狀，則待測物可放至最大，可得到待測物在模型中最大的佔有率，此時即可推估待測物是屬該模型，而達到辨識的效果。然而，由於通常所取得之汽車待測物並非完整的三維物體，因此雖然經過力的推斥計算仍無法達到最正確的位置，也可能無法達到理想的佔有率。所以本論文除了將此人工力場模型實現於真實汽車的車型辨識，也將最小投影矩形及分析待測物之點雲高度的觀念導入，使得辨識時將待測物調整至與 x-y-z平行的位置，可使佔有率得到改善，而提升辨識可靠度。根據這個目標，本論文提出三種改善待測物位置的方法，分別是最小矩形作為待測物之正確位置之估計，點雲高度數量分析找出引擎蓋及車頂，使車體處於與模型平行的位置，而得到合理的車體辨識結果，實驗結果顯示本論文的方法可以提升辨識率而得到正確的車型識別。

In this thesis, the artificial potential field is used as the foundation for matching point cloud with 3D vehicular shapes. An initially small input object (point cloud) is placed inside shape models will experience repulsive force and torque arising from the potential field. A better match in shape between the shape model and the input object can be obtained if the input object translates and reorients itself to reduce the potential while growing in size. The shape model which allows the maximum growth of the input object corresponds to the best match and thus represents the shape of the input object. Since the point cloud does not cover the complete 3D object, the best match cannot be determined only by the potential model. Therefore, the research of this thesis also considers the minimum bounding boxes of projections of the point cloud and the distribution of objects points in the z-direction to improve the matching results. Based on the above concept, three methods are considered to refine the shape matching results via silhouette minimization, hood finding and roof finding. Experiment results show that the methods proposed in this thesis can indeed increase the rate of recognition of real vehicular shapes.