運用多台KINECT對汽車周遭做立體監控

Abstract

本研究提出數項利用架設在視訊監控車周圍的KINECT裝置，對汽車周遭環境做立體監控的方法。該等方法運用KINECT裝置提供的3D感測技術建立環車物體模型，再藉由這些模型開發環車監控系統，用於提醒駕駛人車遭狀況，避免車禍意外發生。 首先，基於針孔成像原理，本研究利用座標轉換將深度影像轉換至3D空間，之後把KINECT裝置取得的深度與彩色影像統整成一3D影像，同時採用幾何方法修正3D影像的彎曲現象。接著，基於遞迴最近點(iterative closest point, ICP)的觀念提出一校正KINECT裝置的方法，將各個環車KINECT裝置的3D影像接合成一完整的環車立體影像。另外，使用從二維的距離加權相關係數(2D distance-weighted correlation, 2D DWC)延伸出來的3D DWC，計算出3D影像間的位移，完整地建立從旁經過汽車的模型，並還原車禍事故現場。 最後，提出一對於路面斜坡和限高障礙物做監控和計算參數的方法，該方法使用三角函數和幾何數學，測量路面斜坡的坡度，並推算車頂和限高障礙物的高度差，以預先提醒駕駛人避免碰撞。另為增進道路駕駛安全，亦提出一針對對向汽車預測可能行進方向與移動的方法。 上列方法的實驗結果良好，顯示出本研究所提系統確實可行。

ABSTRACT In this study, several methods are proposed for 3D monitoring of car surrounds using multiple KINECT devices affixed around a video surveillance vehicle. Around-car object models are constructed with 3D sensing techniques provided by KINECT devices. Based on the constructed models, an around-car monitoring system for car-accident warning and avoidance is developed. Firstly, a method for coordinate conversion based on the pinhole camera model is proposed, which converts the coordinates of the depth image into 3D space points. The method is then used to construct 3D images from the color and depth images acquired with KINECT devices. Also, a geometric correction scheme is employed to correct the bending phenomenon of constructed 3D images. Next, a method for KINECT device calibration using the iterative closest point (ICP) algorithm is proposed, which is used to stitch the 3D images corresponding to the around-car KINECT devices into an around-car 3D image. And by the use of the 3D DWC measure which is an extension of the 2D DWC, the displacements between the 3D images of neighboring KINECT devices are computed, by which a more complete by-passing car model can be constructed and car-accident scenes can be restored. Furthermore, a method for monitoring and computing parameters related to ramps and height-restricting barriers is proposed, which uses trigonometric functions and mathematical geometry to measure the slopes of ramps and calculate the height difference between the car roof and the barrier for pre-warning of possible car collisions. Finally, in order to increase driving safety, a method for predicting the possible direction and movement of the opposite car is proposed. Good experimental results are also shown, which prove the feasibility of the proposed methods for real applications.