利用物體在實際環境與影像空間的資訊調整 擷取頻率之分析與控制

Abstract

近年來，三維定位的相關應用已被廣泛的研究。在過去，影像的擷取頻率並沒有調變的機制，因此，在長時間下因為影像的高擷取頻率，常使得即時三維定位的計算量以及所耗費在儲存影像的記憶體空間變的非常龐大。然而，當物體保持靜止或以非常緩慢的速度移動時，其以高速頻率擷取影像之三維定位的結果與以低速頻率擷取影像之三維定位的結果相去不遠。 本文提出了三種調整影像擷取頻率的控制方法。這三種頻率控制方法皆以物體速度的計算結果去調整影像擷取的時機。在第一種控制方法中(Case I)，影像擷取的頻率是根據物體在真實世界中實際速度的資訊來作調整。在Case I中，所有的照相機會同時調到相同的影像擷取頻率。在第二種控制方法中(Case II)， 影像擷取的頻率是根據物體在影像中速度的資訊來作調整。在Case II中，每一台照相機皆根據自己拍得的影像資訊去調整自己的影像擷取頻率。第三種控制方法(Case III)可以被視作是Case I與Case II的結合，因為它也是以物體在影像中的速度資訊來調整影像擷取頻率，而且會同時把所有的相機調整到相同的影像擷取頻率。 根據本研究中提出的頻率控制方法，在有限的影像儲存空間下，對於目標場景可以記錄的時間會比單純使用高影像擷取頻率來的長。在過去的三維定位應用中，定位的結果是藉由計算每一張影像所得來的。然而，根據本研究提出的頻率控制方法，沒有紀錄到的部分可用一些數學方法來預測。因此，也節省了三維定位的計算量。 本論文利用了三維定位的實驗過程來比較這三套控制方法。在接下來的章節中將會介紹本論文中三維定位過程裡用到的影像處理步驟以及影像分析的方法。 最後，這三種控制方法的實驗結果將會被分析與比較。

In recent years, applications for 3-D positioning have attracted much attention. In previous applications, there is no strategy to adjust image acquisition rate, therefore, the computational load of real-time 3-D positioning and the memory for storing images in a long time will cost much when images are just captured by a higher rate. However, when objects just keep in a static state or move very slowly, the results of 3-D positioning by higher rate are just nearly the same with the lower rate. This thesis presents three control methods for adjusting image acquisition rate. All these three rate control methods adjust the timing for image acquisition by the calculated results of object velocity. The first control method (Case I) adjusts the image acquisition rate by the information of object velocity in real world. In Case I, all cameras are adjusted to a same image acquisition rate synchronously. The second control method (Case II) adjusts the image acquisition rate by the information of object velocity in image. In Case II, each camera adjusts its image acquisition rate based on its own by the information from its captured images. The third control method (Case III) which can be regarded as a combination of (Case I) and (Case II), also adjusts the image acquisition rate by the information of object velocity in image and let all cameras adjust to same image acquisition rate synchronously. In a limited memory for image storage, according to the rate control methods that proposed in this study, the recorded duration for desired scenario is longer than that with purely high image acquisition rate. In previous 3-D positioning applications, the positioning results are obtained one image by one image. However, according to the rate control methods proposed in this study, the 3-D positioning results of the moments which are not recorded can be predicted by some mathematical methods. Therefore, the computational load for 3-D positioning process can be saved. The thesis also presents the experiments of 3-D positioning processes to compare the three control methods. The image processing steps and methods of image analysis for the 3-D positioning processes used in this thesis will be introduced in the fallowing chapters. At last, experimental results for the three control methods are analyzed and compared.