基於雙攝影機之空中手寫數字辨識

由於硬體設備的發展，電腦視覺領域得以將人工智慧技術有效地加以應用。本研究的目的是進行雙攝影機立體視覺下的手指追蹤，並達成空中手寫數字辨識。本研究以雙攝影機產生立體視覺以及深度資訊影像協助手指追蹤，透過深度資訊，能夠判斷手指與鏡頭之間的距離，並排除影像中非屬於此距離範圍中的物體及背景，利用連續密度函數（Probability Density Function, PDF）計算出感興趣區域的深度資訊範圍，並利用粒子群演算法（Particle Swarm Optimization, PSO）進行目標物追蹤（手掌）。有了手掌位置後，找出手指尖的位置及改進深度影像無法檢測遠處細緻物體的問題。接著本系統擷取手指位置、移動路徑、判斷實筆、虛筆並修正軌跡，最後將書寫軌跡輸入至類神經網路做數字的辨別。

關鍵字

人工智慧；空中手寫；深度資訊；實虛筆；雙攝影機

並列摘要

Due to the rapid development of hardware devices, big data analysis and artificial intelligence can be applied into more and more applications. In the field of computer vision, artificial intelligence technology can be effectively applied because of better hardware device support and greater computing power, so problems can be solved more effectively and accuracy can be improved. The focus of this study is to perform finger tracking under dual-camera stereo vision and achieve air-handwritten digit recognition. This study uses a dual camera to generate stereoscopic and depth information images to assist finger tracking. Through depth information, the distance between the finger and the camera can be determined, and the objects and background in the image that are not within this distance range are excluded. The proposed system uses the probability density function (PDF) to calculate the depth information range of the region of interest, and uses particle swarm optimization (PSO) to track the target (palm). With the palm position, this system further analyzes the position of the tip of the finger and improves the depth image cannot detect the problem of distant objects. Then the proposed system extracts the position of the finger, moves the path, determines the real and virtual strokes and corrects the trajectory, and finally inputs the writing trajectory to the artificial neural network to make a digital recognition.