Have library access?
  • Journals
  • OpenAccess




台灣社會人口老化及少子化已造成農村勞力不足的問題,並已影響到農業正常的發展。台灣農業正逐漸轉型中,將機器人應用到農業領域是一種趨勢,可以舒緩農業缺工的問題,且可以執行原本人類不易處理的工作。本研究的目的是建立一套番茄採收機器人視覺系統,該系統搭配研究合作團隊之宜蘭大學生機系所開發的爪具設備系統,開發完成應用於溫室設施內之水果採收機器人。建立之機械視覺系統包含了三部份:影像擷取、影像處理與分析、雙眼立體視覺系統。將擷取之影像依當時之光環境狀況以對應的四個方程式計算其最佳影像擷取條件,包含曝光時間及放大倍率。另外,採用對數型迴歸白平衡演算法校正影像顏色,使系統擷取影像在3~5秒內完成,並獲得良好的影像品質。影像處理則利用Hue色層將番茄挑選出來,並搜尋其外部特徵位置資料(二維座標)。使用單支攝影機搭配移動平台,進行雙眼視覺系統的建立,獲得深度資料(第三維座標),提供給採摘爪具及行走載具進行其路徑規劃。雙眼視覺系統的性能測試,發現在60 cm~80 cm的工作距離範圍下,XY座標平面距離誤差小於1.5 mm,Z軸距離誤差小於4.4 mm。在水果目標物重疊率25,50與75%的情況下,判斷正確率分別為100,85.2與55.6%。最後將本機械視覺系統整合爪具設備系統進行試驗,研究結果顯示,直徑60,70與80 mm的球體,採摘成功率分別為93.3,91.1與84.4%。研究結果顯示本研究開發成功的視覺系統能成功擷取良好的番茄影像資訊,並達成番茄採收機器人作業之要求;本研究的成果可以進一步應用至其他水果之機器人採收作業。

Parallel abstracts

Population decline and aging cause serious problems for labor shortage in agriculture, and it has affected the future development of agriculture in Taiwan. It is a trend to apply robots to agriculture. The aim of this study is to develop a machine vision system for robotic harvesting of tomatoes. That system will be integrated with the end effector (claw device) system, which was built by partner research team, I-Lan University, to develop a fruit harvesting robot in greenhouses. The developed machine vision system in this study was composed of three parts, image grabbing, image processing, and stereo vision system. The best image capture conditions including exposure time and gain value were determined by developed corresponding four equations which were in accordance with the on-site lighting conditions. Log-regression-white-balance method was also adopted to calibrate the color of captured images. The system could capture a high quality image in 3~5 seconds. Hue index from HIS color system was selected for image processing to identify the targeted tomatoes. Two dimensional data of tomato positions were thus specified. Stereo vision system was developed by using a single camera with a mobile slide platform to measure the third dimensional coordinate of the tomato. Three dimensional data of tomato positions would then be sent to end effector system and the carrier to plan the paths for tomato picking. The performance of stereo vision system has been evaluated and obtained an excellent result, within the working distance range of 60~80 cm apart from system, the results showed the error was less than 1.5 mm in X and Y directions, and less than 4.4 mm in Z direction. The discrimination accuracy was 100, 85.2 and 55.6% when the overlapping portion of objects was 25, 50 and 75%. Regarding the performance of integrated robotic system consisting of machine vision and end effector systems, the picking accuracy was 93.3, 91.1 and 84.4% respectively when object diameters were 60, 70 and 80 cm. The results indicate that the developed machine vision system meets the requirements of tomato picking robot, and has the potential to apply to robotic harvesting for other fruits.