本文選題：關鍵幀提取 + 背景運動補償��； 參考：《東北電力大學》2017年碩士論文

【摘要】：我國經濟正在飛速發(fā)展,國內對于電力的需求也急速增長,高壓線是電力供應的保障,但是高壓線常年暴露于野外環(huán)境,常有異物懸掛在輸電線路上,導致墜線的危害,嚴重影響居民的正常生活和企業(yè)的安全生產。傳統人工巡線的方式耗時耗力,并且受到地理環(huán)境和氣候變化的嚴重制約;然而飛行器巡線可以在遠距離和復雜環(huán)境中準確地進行電力安全檢測,該技術集飛行器控制和圖像處理技術為一體,使用機載攝像頭收集航拍視頻,并通過圖像技術分析航拍視頻,從而檢測高壓線的運行狀態(tài),飛行器航拍方式能夠自動檢測線路運行狀態(tài),同時有效提高異物檢測的準確性。本文針對輸電線路懸掛異物的問題,采用目標檢測技術分析輸電線路的航拍視頻,達到自動標注并準確跟蹤異物的目的。首先提取輸電線路航拍視頻的關鍵幀,依據視頻中背景信息復雜但是變化緩慢的特點,采用漸變鏡頭邊緣檢測算法提取關鍵幀,通過預估漂移區(qū)域法改進雙閾值鏡頭檢測算法,達到保留視頻中關鍵信息的目的;其次在背景補償方面,使用投影法估計背景的運動量,采用預測器來改進投影法,達到快速并準確估計背景運動量的目的,從而彌補相鄰幀間背景的偏差量;再次使用三幀差法檢測異物,使用二維交叉熵法改進閾值分割技術,并通過概率密度函數精簡異物區(qū)域,利用八連通區(qū)域法標注異物區(qū)域,達到準確檢測并標注異物區(qū)域的目的;最終通過提取ORB特征點跟蹤輸電線路異物,以參考幀中異物區(qū)域為基礎,使用Mean-Shift算子預估當前幀中異物區(qū)域,提取兩幀中異物區(qū)域的ORB特征點,通過PROMATCH算子改進漢明距離,達到準確跟蹤輸電線路異物的目的。本文采用預估漂移區(qū)域和雙閾值法結合來檢測關鍵幀,能夠有效剔除視頻中冗余幀,在保證所選幀中包含主要內容的同時,減少系統計算量;在背景運動補償中采用最小二乘預測器,可以改變灰度投影法逐幀搜索的弊端,提高背景運動量的計算效率;在運動目標檢測方面使用二維交叉熵閾值法,能夠增加異物區(qū)域的完整度,并且減少非異物區(qū)域的影響;在異物跟蹤中采用ORB算子,有效縮短特征點間匹配時間。由實驗分析可知,本文可實現異物的檢測、跟蹤操作。
[Abstract]:China's economy is developing rapidly, and the domestic demand for electricity is also growing rapidly. The high-voltage line is the guarantee of power supply, but the high-voltage line is exposed to the field environment all the year round, and foreign bodies are often hung on the transmission line, which results in the harm of falling line. Seriously affect the normal life of residents and safe production of enterprises. The traditional manual inspection method is time-consuming and labor-intensive, and is severely restricted by geographical environment and climate change. However, aircraft patrol can accurately detect electric power security in long distance and complex environment. The technology integrates the aircraft control and image processing technology, uses the airborne camera to collect aerial video, and analyzes the aerial video through the image technology, so as to detect the running state of the high-voltage line. Aerial photography can automatically detect the running state of the line and improve the accuracy of foreign body detection. In this paper, aiming at the problem of transmission line hanging foreign body, the target detection technology is used to analyze the aerial video of transmission line, and the purpose of automatic marking and accurate tracking of foreign body is achieved. Firstly, the key frame of transmission line aerial video is extracted. According to the characteristics of complex background information but slow change in the video, the edge detection algorithm of gradient shot is used to extract the key frame, and the dual threshold shot detection algorithm is improved by predictive drift region method. Secondly, in the aspect of background compensation, the projective method is used to estimate the motion of the background, and the projector is used to improve the projection method, so that the background motion can be estimated quickly and accurately. In order to make up for the deviation between adjacent frames, three frame difference method is used to detect foreign body again, two dimensional cross entropy method is used to improve threshold segmentation technology, and the probability density function is used to simplify foreign body region, and eight connected region method is used to mark foreign body region. Finally, by extracting the ORB feature points to track the foreign bodies in the transmission line, and based on the foreign body region in the reference frame, the foreign body region in the current frame is estimated by using the Mean-Shift operator. The ORB feature points of the foreign body region in the two frames are extracted, and the hamming distance is improved by the PROMATCH operator to track the foreign body in the transmission line accurately. In this paper, the combination of predictive drift region and double threshold method is used to detect the key frames, which can effectively eliminate the redundant frames in the video, while ensuring that the selected frames contain the main contents while reducing the system computation. In background motion compensation, the least square predictor can change the disadvantage of the gray projection method to search frame by frame, and improve the computational efficiency of background motion, and the two-dimensional cross entropy threshold method is used in moving object detection. It can increase the integrity of the foreign body region and reduce the influence of the non-foreign body region, and the ORB operator can effectively shorten the matching time between the feature points in the foreign body tracking. According to the experimental analysis, the detection and tracking of foreign bodies can be realized in this paper.
【學位授予單位】：東北電力大學
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：TP391.41;TM755

【參考文獻】

相關期刊論文 前10條

1 曾巧玲;文貢堅;;運動目標跟蹤綜述[J];重慶理工大學學報(自然科學);2016年07期

2 鐘_，

本文編號：1969737