應用智慧型雙眼機器人於物體運動偵測、預測與狀態描述之研究(I)

Abstract

本計劃嘗試開發智慧型雙眼機器人(eyeRobot)，以視覺為基礎執行物體運動偵測、預測與狀態描述，展開為期兩年的研究。目前計畫主持人已經開發了影像處理技術、學習與灰預測法則、影像與控制硬體實現技術、雙眼機器人，分別如下所示： 1. 仿人類眼球運動機器人機構設計。 2. 影像縫合技術。 3. 應用於無背景限制下三維建模之物體萃取演算法。 4. 演化策略為基礎的學習控制系統。 5. 以灰預測設計離散型順滑模態控制器。 6. 以CAM為基礎之樣式累加向量法在車牌字元辨識系統之應用 7. 以DSP為基礎實現馬達控制。 由於影像的技術需要大量的數學運算時間，因此本計劃改以類神經網路為主要架構，讓eyeRobot利用學習及灰預測的智慧型法則，學習疊合雙眼影像並建構立體視知覺，此外也將控制技術直接實現於數位訊號處理器上。本計劃預計兩年內可達成以下之目標： 第一年：以類神經網路為基礎，學習雙眼影像疊合技術與建構立體視知覺，以偵測物體。 第二年：以類神經網路學習以及灰預測智慧型法則為基礎，預測及描述物體運動狀態，提供物體間在空間與時間上的相對關係，並以數位訊號處理器實現控制技術，以追蹤運動物體。 完成以上目標之eyeRobot將可應用於各類型的監控系統，主動提供監控地點的物體運動狀態。

This project proposes an intelligent binocular robot, or called eyeRobot, to detect, predict and describe the motion of objects in two years. Recently, some related key technologies have been devloped and listed as below: 1. Machanism design of human-like eye movement robot. 2. Image stitching technology. 3. A novel object extraction algorithm in 3D model reconstruction for objects in arbitrary background. 4. Evolution strategies-based learning control system. 5. Discrete time sliding-mode control design with grey predictor. 6. A CAM-based license plate character recognition system using the pattern accumulated vector method. 7. DSP-based motor control. Since general image stitching requires a large amount of computation time, this project instead adopts the neural network for the stitching of binocular images, based on the training process, learning algorithms, and prediction. Furthermore, the control of eyeRobot would be implemented on the hardware of DSP board. This project would take two years to achieve the following goals: 1st year: In order to detect objects, the eyeRobot would learn to stitch binocular images and construct stereo visual perception based on neural networks. 2nd year:In order to acquire the spatial-temporal relationship of objects, the eyeRobot would be developed to predict and describe the motion based on learning in neural networks and prediction with grey theory. Furthermore, the control technology to track moving objects would be implemented on the DSP board. After this project, the developed eyeRobot would be applied to monitoring system that effectively and efficiently provides the information of the motion of objects under supervision.