在人機互動系統中，觸控手勢操作已成為新一代產品的必備功能。目前觸控手勢辨識主要透過後端中介軟體運算實現，這無疑加重系統工作負擔，非常不適合要求即時且資源有限的嵌入式系統使用。本研究提出一個即時手勢辨識演算法，首先對手勢操作的連續座標進行正規化處理，接著透過機率神經網路(Probability Neural Network, PNN)進行手勢辨識與分類，並將粒子群體最佳化(Particle Swarm Optimization, PSO)演算法應用於機率神經網路的平滑參數σ最佳化，改善手勢辨識率。我們以MIAT方法論將即時手勢辨識演算法全硬體化為一個智慧型觸控控制器，於FPGA平台進行硬體驗證。使用管線化(Pipeline)平行架構重新設計硬體電路以提高系統效能，並加入使用者自訂手勢功能及運用錨點概念來模擬實現多點觸控手勢功能。實驗結果顯示，此智慧型觸控控制器具有高效能、低功耗及高辨識率的特性，大幅簡化觸控應用系統設計，可適用於更小更省電的可攜式電子產品。 In human-computer interaction systems, touch gestures have become the must-have feature of the new generation products. At present, touch gesture recognition is mainly implemented through the middleware of back-end system, which undoubtedly increase the system workload and is not suitable for real-time requirements and limited resources of embedded systems. This paper proposes a real-time gesture recognition algorithm. We first normalize the continuous coordinates of gestures, and then use Probability Neural Network (PNN) for gesture recognition and classification, and Particle Swarm Optimization (PSO) algorithm for PNN, which optimize the smoothing parameter σ, to improve the rate of gesture recognition. The algorithm is then synthesized into a smart touch controller by the MIAT hardware synthesis methodology and verified on a FPGA platform. We redesign the hardware circuit by using pipelined parallel architecture to improve system performance, add user-defined gestures and anchor point to simulate the multi-touch gestures. From the experiments, the smart touch controller has high performance, low power consumption and high recognition rate, and significantly simplifies the design of touch application system, which is applicable for smaller and more power saving portable electronic products.
