移動式人流計數系統之開發與研究

Design and Implementation of a Portable People Counting System

以往在需要統計人流的時候，我們必須使用人工的方式，在一個定點統計所經過的人數，這樣不僅會因為人計算上的失誤，而導致誤差，更會耗費大量的人力成本。隨著科技的進步，出現許多自動化人流計數系統，並且發展出各自解決方案，但這些系統需要在固定定點架設，無法隨意的移動。且這些系統都需要由專業人員來進行架設與調整才能提供預期的計數效果。藉由現在行動裝置移動的方便性，提供利用行動裝置來進行人流計數，可以隨意的在任何地點使用，也不需專業人員來進行架設。我們的研究提出針對垂直、水平及傾斜角度所拍攝影片進行人流計數，並且針對行動裝置進行最佳化，使得人流計數在行動裝置上可以順利的被運用。最後經由實際測試我們在垂直拍攝方向影片上獲得約89%的準確率，水平拍攝角度則有78%的準確率，而傾斜拍攝角度則獲得72%的準確率。所以我們所提出的方法已經可以運行於行動裝置上，並且提供一定的準確率。

People counting has been an important component in video surveillance systems. Video cameras are usually installed at fixed locations by professionals, and the recorded videos are processed to count the people flow either by human labors or image processing systems. As mobile technology advances, it is now possible to easily set up surveillance cameras at anywhere to get the information of people flow for a certain location by just using general mobile devices, which can be used to assess the feasibility of a business model, such as the opening of a chain store. In this paper we present the design of a portable people counting system which is implemented in an android-based mobile device. The system consists three subsystems which allow videos to be taken from three different angles: vertical, horizon, and inclination. Our experimental results show that the average accuracies of vertical, horizon, and inclination shots reach as high as 89%, 78%, and 72%, respectively.