篇名 | 嵌入式寬角度即時行人偵測與移動預測技術 |
---|---|
卷期 | 151 |
並列篇名 | Real-time Embedded Pedestrian Detection and Motion Prediction Techniques for Wide View Camera |
作者 | 黃鐘賢 、 李宗展 |
頁次 | 079-086 |
關鍵字 | 嵌入式系統 、 行人偵測 、 物件追蹤 、 行動預測 、 Embedded System 、 Pedestrian Detection 、 Object Tracking 、 Motion Prediction |
出刊日期 | 201306 |
本技術使用由複數台攝影機所組成之寬角度取像模組,於嵌入式平台上開發車用行人偵測與行人追蹤技術,主動偵測視野內之行人,有效提醒駕駛人,提升行車安全。然而行人偵測演算法的運算需求極高,主因在於1.攝影機解析度越高,需要執行偵測的次數越多;2.偵測準確率越高,每次偵測的運算量越高。為了使行人偵測演算法可運行於嵌入式平台,本技術研發一基於攝影機幾何及影像能量導引之快速行人偵測技術,能有效提高偵測速度7~10倍,運作於嵌入式平台達可達每秒20幀以上。
In this research, a pedestrian detection and tracking technology is developed especially for embedded systems equipped in vehicle. The image capture module with wide view is composed of multiple cameras. The pedestrians in the view are automatically detected to inform the driver. However, the computations of traditional pedestrian detection algorithm are very huge. The reasons are two: 1. The higher the resolution of camera is, the times of detection are more; 2. The higher the correct detection rate, the computations of detection are higher. In order to perform pedestrian detection on an embedded system, we developed a fast pedestrian detection technology based on camera geometry and image energy guided. Its execution rate is 7~10 times than traditional algorithm and up to 20 frames per second (fps) on our experimental embedded platform.