<人在做, GPU在看> 專為追蹤人群所開發的GPU高速影像檢測 -- 麻省理工學院MIT碩士論文

試想一下, 每日動輒十萬陌生人次流量的機場, 如何追蹤紀錄每一個人的行走路徑?

GPU高速人群追蹤, 不再只是針對有前科紀錄的臉譜. 機場無數的攝影鏡頭, 可以在瞬間把無數陌生人的臉譜特徵分析/紀錄+追蹤+繪製行走路徑.

根據無數的繪製路徑, 若發現陌生人(即使沒有前科紀錄)同一天多次進出機場, 或者多次在兩個航廈大樓之間反覆移動, 先進的GPU高速人群追蹤系統可以報警, 並將疑點告知航廈安全管理員.

人在做, GPU在看

專為追蹤人群所開發的高速影像檢測 -- 麻省理工學院MIT碩士論文
Fast Human Detection with Cascaded Ensembles
(Master's thesis submission, Dept. of Electrical Engineering and Computer Science, MIT)

本論文探討影像檢測與判讀上的困境, 尤其是人類影像追蹤領域. 本論文研究團隊採用NVIDIA CUDA, 一種已經被認可為業界標準的平行運算架構, 來處理通用型C語言CUDA高等運算(也就是俗稱的”C for CUDA”), 充分發揮可程式化超多核心GPU處理器的優勢, 達到人群追蹤領域上, 可觀的效能暴衝.
Extract: "This thesis addresses the problem of object detection from images, in particular the detection of people. As digital cameras become more widespread, the volume of available data to digital camera owners reaches such a point that digital content management presents itself as a problem. In our work, we use the NVIDIA CUDA framework. CUDA is the computing platform that enables developers to code parallel algorithms through industry standard languages. The CUDA programming model acts as a platform for massively parallel high performance computing by providing a direct, general-purpose C language interface (‘C for CUDA’) to the programmable multiprocessors on the GPUs. When implemented on this platform, we observed a significant speed up in our cascade detector‘s performance." Authored by Berkin Bilgic, MIT student. See: http://is.gd/bXKsK.