全周俯瞰監視與側邊偵測系統

Surrounding Top-view Monitor and Lateral Detection System

李國煒(Kuo-wei Li)

指導教授 ： 曾定章

國立中央大學/資訊電機學院/資訊工程學系/碩士(2011年)

摘要

道路交通事故的部分因素是因為車輛行進中沒有看到障礙物而發生的，尤其是車體結構與後照鏡角度造成的盲點區域，使得駕駛無法了解車輛週遭環境而造成人員與車輛損傷。為避免看不到車輛週遭環境而造成的交通意外，並提高停車時的安全性。我們提出一套全周俯瞰監視與偵測系統，並將之實現於DSP嵌入式系統中。整個系統共包含兩大部份：一是全周俯瞰監視用於輔助駕駛監視車輛周遭的狀況，二是側邊偵測用於協助駕駛主動偵測車輛周遭的障礙物。
全周俯瞰監視與偵測系統在車輛四周架設廣角相機以拍攝車輛週遭影像，經過離線處理扭曲校正、暗角消除、俯瞰轉換後，得到四周俯瞰影像的相對關係。再使用一部相機由上方拍攝車輛四周的特徵，將俯瞰影像快速對位為一張俯視車輛週遭的全周俯瞰影像，最後將各項參數建立一張查找表，在線上處理階段根據查找表查表內插與校正影像。動態側邊偵測系統則是以側邊影像估計光流，藉由光流濾除及群聚後，擷取障礙物主動提示駕駛者。
嵌入式全周俯瞰監視系統可在影像的解析度為720 × 480的情況下，於Texas Instruments? DaVinci™ DM648 900 MHz Digital Media Processor開發板上執行可達每秒10張的處理速度。而側邊障礙物偵測程序可在影像顯示大小為320 × 240的情況下，在Intel Core™2 Duo 2.83GHz及1.99GB RAM的個人電腦上可達每秒22張，障礙物偵測率可達94%。

關鍵字

影像對位 ； 影像對位 ； 鳥瞰轉換 ； 影像扭曲校正 ； 影像暗角補償 ； 相機參數校正 ； 光流 ； 障礙物偵測

並列摘要

Partial traffic accidents are resulted from drivers can’t watch the whole vehicle surroundings. To reduce the accidents caused by collision of surrounding obstacles, we mount four wide-angle cameras at the front, rear, and both lateral of the vehicle to capture consecutive images; then we present a real-time surrounding top-view monitor and a lateral obstacle detection system for parking assistance.
In offline steps of surrounding top-view monitor system, we first estimate camera intrinsic and extrinsic parameters, and also calibrate the parameters of distortion model and vignetting model for distortion correction and vignetting compensation. Then we calibrate the geometric relationships of four cameras using a proposed multi-camera calibration method. Third, we calculate the feathering weights of pixels to produce a seamless surrounding top-view image. At last, we build lookup tables for recording the mapping between the captured images and the surrounding synthesized image to speed up the processing. After offline steps, our system online interpolate and calibrate the surrounding synthesized image by those lookup tables directly.
In lateral obstacle detection system, we utilize the calibrated lateral images to estimate the optical flow of possible obstacles. Then we filter and group the non-ground optical flow by direction of motion vectors and color of feature pixels. Third, we determine whether the optical-flow groups are obstacle or not. Finally, the detection system will alarm if there is an obstacle to be collided by the vehicle. In our experiment, the system detection rate is about 94%.

並列關鍵字

obstacle detection ； camera calibratio1n ； optical flow ； distortion correction ； vignetting compensation ； homography ； image stitching ； color blending

參考文獻

[1] Bertozzi, M. and A. Broggi, "GOLD: a parallel real-time stereo vision system for generic obstacle and lane detection," IEEE Trans. on Image Processing, vol.7, no.1, pp.62-81, 1998.

[4] Devernay, F. and O. Faugeras, "Straight lines have to be straight," Machine Vision and Applications, vol.13, no.1, pp.14-24, 2001.

[7] Enkelmann, W., "Obstacle detection by evaluation of optical flow fields from image sequences," Image and Vision Computing, vol.9, no.3, pp.160-168, 1991.

[9] Gandhi, T. and M. M. Trivedi, "Parametric ego-motion estimation for vehicle surround analysis using an omnidirectional camera," Machine Vision and Applications, vol.16, no.2, pp.85-95, 2005.

[10] Gandhi, T. and M. M. Trivedi, "Vehicle surround capture: survey of techniques and a novel omni-video-based approach for dynamic panoramic surround maps," IEEE Trans. on Intelligent Transportation Systems, vol.7, no.3, pp.293-308, 2006.