應用光學影像於週期擺動機器魚之定位研究

Application of Optical Images on Localizing a Periodically Oscillating Robotic Fish

楊皓甯(Hao-Ning Young)

指導教授 ： 郭振華

國立臺灣大學/工學院/工程科學及海洋工程學研究所/碩士(2020年)

https://doi.org/10.6342/NTU202000464

摘要

仿生型自主式水下載具常應用於群體追蹤任務之中，辨識有特定擺動頻率之目標物可用於載具間之追蹤。本論文建立了一套智慧型即時性光學辨識系統，可以在水下環境利用雙眼視覺辨識特定週期擺動的物體，並求其頻率、定位及測距。在背景環境單純下，用靜態動態分離演算法辨識有特定週期擺動運動的仿生型自主式水下載具。而在求其頻率程序中，先調整好適當的幀率後，用快速傅立葉轉換求得擺動物體之頻率。
定位測距使用了立體匹配演算法；Semi-Global Block Matching(SGBM)演算法以及粒子濾波器演算法去估測精確的定位與測距，以使追蹤者在用雙眼視覺追蹤載具同時，藉由測距資料來設定彼此之間的安全距離以避免碰撞。本文最後將此定位測距演算法運算之結果與雷射掃描儀所量測之誤差比較，以驗證此光學系統在測距功能上之可行性。最後，經由機器魚的水槽實驗，驗證上述系統功能上之可行性。

關鍵字

仿生型自主式水下載具 ； 機器視覺 ； 靜態動態分離演算法 ； 立體匹配演算法 ； 序列式蒙特卡羅辨識演算法

並列摘要

Biomimetic autonomous underwater vehicles (BAUVs) often used in group tracking task, so recognizing the target with specific period is inevitable for this task. This work builds an intelligent optical real-time recognition system for this kind of application. The function of the established system is involved in recognizing target with specific periodic motion by utilizing with stereo cameras, obtaining the frequency and estimating the distance of it in the underwater environment. In this paper, in the simple background environment, as for identifying the BAUV with specific periodic motion, we use the mixture-based background/foreground segmentation algorithm. And in the procedure of obtaining the frequency of target, we use the fast Fourier transform to achieve it before setting the appropriate framing rate.
As for the estimating distance, we use the semi-global block matching (SGBM) algorithm. The purpose of it is to maintain the safe relative distance between them which is set by distance data to avoid tracking crash as tracker pursuit the object. And in my paper, we present a novel real-time optimizing estimation method, named the sequence Monte-Carlo method to estimate the higher accurate distance data. In particular, the augmented reality technique is involved to the estimated updated model. Then, we do the error comparison on the estimating distance by utilizing with the algorithm of system and laser scanner to verify the feasibility of the function of optical system related to estimating distance. Finally, above functions of system are verified by experiments in a water tank.

並列關鍵字

biomimtic autonomous underwater vehicle ； machine vision ； mixture-based background/foreground segmentation algorithm ； SGBM algorithm ； Sequential Monte Carlo recognition algorithm

參考文獻

[1] S. Shah, J. K. Aggarwal, “Depth Estimation Using Stereo Fisheye Lenses,” IEEE International Conference on Image Processing, Vol. 2, pp. 740-744, 1994.

[2] Yang Gao, Xiquan Fu, Yanfeng Bai, “Ghost imaging in transparent liquid,” Journal of Optics, Vol. 46, Issue. 4, pp. 410-414, 2017.

[3] Mingnan Le, Gao Wang, Huabin Zheng, Jianbin Liu, Yu Zhou, Zhou Xu, “Underwater computing ghost imaging,” Optics Express, Vol. 25, Issue. 19, pp. 22859-22868, 2017.

[4] L. A. Marcomini, A. L. Cunha, “A Comparison between Background Modeling Methods for Vehicle Segmentation Highway Traffic Videos,” Computer Science Bibliography, 2018.

[5] Chris Snyder, “Particle filters, the “optimal” proposal and high-dimensional systems,” Seminar on Data assimilation for atmosphere and ocean, 6-9, 2011.