實現即時車輛偵測與距離估計之嵌入式平台

Abstract

嵌入式平台(Embedded platform)是目前很普遍被使用的平台架構，它的低成 本、低耗電特性使之應用層面更為廣泛。而 x86 平台是目前大家所最熟悉的平 台，也因此使用此平台開發軟體的時間會縮短很多。但因現有使用Intel x86 中 央處理器的 x86 平台的大多都很耗電，此論文中所使用之中央處理器是 800MHz 的VIA C3 相對於 Intel 者省電許多，再結合 VIA 開發的嵌入式平台 的機版 EPIA-N 其大小只有12cm×12 cm，突破了以往 x86 平台高耗電與大尺 吋的印象，適合應用在嵌入式系統上。 此論文是利用一個 CMOS 感應器，將擷取到的影像資料，透過影像邊緣偵 測(edge detection)處理後，分別同時交給車道偵測(lane detection)與車輛偵測 (vehicle detection)演算法去做車道與車輛的偵測。給予偵測出之車道與車輛的資 訊，即可利用距離遠近所造成車道在視覺呈現的比例關係推算出與前車的距離。 此架構可應用於煞車系統，當偵測出與前車距離過近時，自動啟動煞車系統， 避免碰撞產生，或是當駕駛過於疲憊時，造成車距過近，提供即時的警告訊息。

Embedded platform is popularly used for its low cost and low power characteristics. Using a familiar x86 platform can shorten the software development schedule. But, the system power consumption is very high and the size too large for most of the x86 systems using an Intel x86 CPU. The platform presented in this Thesis adopts a 800MHz VIA C3 CPU and the system board size is 12cm×12cm. Its low power consumption and very small size make it a good platform for embedded system applications. The proposed embedded platform uses a CMOS sensor to capture the image of a vehicle in front, and transfers it through a USB interface to a vehicle filter application in the embedded platform. After the vehicle filter application transforms this color image to a gray scale image, from which an Absolute Difference Mask (ADM) algorithm [1] is used to extract the image edges for detecting the road lane and vehicle information. For the lane detection, we use a TFALDA algorithm [4]. As to the vehicle detection, we use the vehicle symmetric features to search the existence of a vehicle. When the lane and vehicle information are found, we can use the lane width measured from the vehicle to compute the distance of a front vehicle. For the application side, this platform can be used as a vehicle automatic breaking system, or as a secure-distance keeping or warning system.