結構光源 RGB-D 攝影機校正

RGB-D 攝影機作為一台同時擁有彩色影像與深度影像擷取能力的裝置，有非常大的應用潛力。它可以用來估測肢體動作，提供體感遊戲的操控、估測病人姿態以利照護，亦可輔助機器手臂抓取物體、協助機器人在未知場地上進行室內場景 3D重建，甚至可以做到人臉辨識、 3D 建模等。 RGB-D 在出廠時皆曾做過校正，但經過長時間的運作，其誤差就會越來越大，因此許多學者提出各種方法對深度量測值進行修正。這些現有的方法主要可以分為兩大類，第一類為參數化的校正方法，這類型的校正模型由於過度簡化，無法有效校正誤差；第二類則是使用非參數型的校正方法，這類型校正方法雖然可以達到很高的精確度，但主要的問題就是所需估測的參數量過大。在這篇研究當中，我們認為 RGB-D攝影機的電子元件在運行中發熱，易導致電路板變形，使得深度估測用的 IR 攝影機與紅外光光斑投射器的方位產生變化，是深度誤差隨著使用時間變長而劣化的主因。針對方位變化，我們提出一個線性參數化模型，並發展不需特殊設備的校正方法。這個模型與方法經實驗驗證，可以有效降低深度估測的偏差。

關鍵字

攝影機校正；結構光源； RGB-D攝影機；深度攝影機誤差；深度更正

並列摘要

Due to its ability of simultaneously acquiring both color and depth images of the scene, an RGB-D camera has great application potential. Its built-in function of posture estimation can be used for game controlling, patient care improvement, assisting a robot arm to grasp an object, 3-D scene reconstruction, 3-D object reconstruction, etc. Normally, an RGB-D camera was calibrated in the factory to achieve its designed depth accuracy. However, its accuracy will deteriorate with operating time. Therefore, many methods have been proposed to recalibrate an RGB-D camera. The existing methods can be classified into parametric approaches and non-parametric approaches. However, the existing parametric approaches used overly simplified models so that the depth estimation error cannot be efficiently reduced. Conversely, although the non-parametric approaches can achieve very high calibration accuracy, the number of parameters to be estimated is too large which may hinder its application. In this study, we observe that when an RGB-D camera is operating, the dissipated power of its electronic devices will heat the printed circuit board (PCB) locally and the non-uniform heating will eventually deform the PCB. Furthermore, deformation of the PCB will alter the relative position and orientation between the depth camera and the projector. Therefore, we assume that deformation of the PCB of a structured-light RGB-D camera is the main reason for depth accuracy deterioration. Based on this assumption, we propose a new parametric method for RGB-D camera recalibration. The proposed method uses a linear model with parameters related to the small change of the relative position and orientation. It is linear and does not require any special calibration equipment. Real experimental data have confirmed that the proposed method can effectively reduce the depth estimation bias.

並列關鍵字

Camera Calibration ； Structured-Light ； RGB-D Camera ； Depth Camera Error ； Depth Correction

參考文獻

[1] D. Herrera, J. Kannala, and J. Heikkilä, “Joint depth and color camera calibration with distortion correction,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 34, no. 10, pp. 2058–2064, 2012.

Google Scholar

[2] B. Karan, “Calibration of kinecttype rgbd sensors for robotic applications,” FME

Google Scholar

Transactions, vol. 43, pp. 47–54, 2015.

Google Scholar

[3] K. Khoshelham and S. O. Elberink, “Accuracy and resolution of kinect depth data for indoor mapping applications,” Sensors, vol. 12, no. 2, pp. 1437–1454, 2012.

Google Scholar

[4] C. Zhang and Z. Zhang, “Calibration between depth and color sensors for commodity

Google Scholar

國際替代計量

結構光源 RGB-D 攝影機校正

全文下載

主題瀏覽