邵校正: 三維飛時相機深度校正

Abstract

iTOF (indirect Time-of-Flight)間接飛時相機是藉由偵測到光波訊號碰到物體來回的相位差進行深度預測。如：Kinect 2.0、Samsung Galaxy手機深度鏡頭都是使用相關技術。然而，iToF 相機在偵測深度的過程中非常容易被外界影響，如：光線、溫度、接收器上每一個像素所接收到的時間差等都是需要在iToF相機出廠前進行校正以確保深度精準度。此篇研究主要探討iToF系统性的錯誤：擺動(Wiggling)、固定相位模式雜訊 (Fixed Phase Pattern Noise)、溫度漂(Temperature Drift)、鏡頭校正(Lens Distortion)，四個項目進行深度鏡頭校正任務。 此篇論文會首先透過相位差計算偵測點深度，再透過傅立葉轉換找出最高與次高頻區間段，利用傅立葉反轉換回推每一像素針對擺動誤差的修正。固定相位模式雜訊為了將整體誤差降到最低，會透過不同深度計算偵測與實際深度誤差進行評估與多項式函式擬合方法找出最接近數值。此篇論文創造將四種不同深度放於一張相片進行校正，可以在最短時間內找出固定相位模式誤差的參數統計。溫度飄移在本篇論文會透過熱箱在不同溫度下的統計進行查表校正。最後在鏡頭校正部分則使用針孔相機模型與張正友校正法進行鏡頭評估。找出相機內部與外部參數，同時也計算出輻射失真、鏡頭與相機感測器之間的切向失真參數。

iToF, indirect Time-of-Flight, camera predicts depth by detecting the phase difference between the light-wave signal and the object, such as Kinect 2.0, and Samsung Galaxy mobile phone. However, iToF camera will easily be affected by the external environment when detecting depth, for example, sunlight temperature, time difference received by each pixel on the receiver, and so on. All need to be calibrated before leaving the factory to ensure depth accuracy. This study will mainly discuss the systematic errors of iToF: Wiggling, Fixed Phase Pattern Noise, Temperature Drift, Lens Distortion: the four main errors that affect results. This thesis will first discuss how to calculate the depth information through different phases detected by iToF sensor, and then use Fourier transform to find the highest and second-highest frequency intervals, then use inverse Fourier transform to push back the correction of the wiggling error for each pixel. Besides, to minimize the fixed phase pattern noise, the actual depth errors are evaluated through the polynomial function fitting method. To get better results, we also focus on temperature drift by look-up table through the statistics of the thermal chamber at different temperatures. Finally, we use pinhole camera model to calibrate lens distortion.