透過您的圖書館登入
IP:3.238.195.81
  • 學位論文

以巨觀模式為基之異常行為偵測及行為辨識視訊系統

Macro View-based Abnormal Event Detection and Activity Recognition in Video Sequences

指導教授 : 蔡篤銘

摘要


本研究利用機器視覺技術,發展一套智慧型視訊監控與行為辨識系統,以機器視覺為基礎,自動偵測出環境中無法預期的異常事件或辨識特定動作,可用於居家安全監控、老人活動力監測、人機互動及特定異常行為偵測等應用,本論文所提出智慧型視訊系統之演算法包含三個步驟:1) 動態主體分割,2) 時-空動作表達方法,3) 行為分類與辨識,本論文將針對每個步驟,提出對應之演算法,並將所提出之方法應用在異常行為偵測及行為辨識。 在前景分割方面,以眾數為基礎,找出每個像素點出現頻率最高的灰階值作為背景模型,眾數前景分割方法可有效偵測出低對比之主體且計算快速,此前景分割結果可用於後續發展之動作表達方法。在行為表達方面,本研究使用巨觀之策略,針對前景分割結果中移動主體在時間與空間維度所造成的動態變化建構出全域式的表達方式,可同時紀錄每個主體的行為與移動情形,將各種行為表達成一張全域式的動態能量圖,透過此策略可免除現有行為偵測需進行前景肢體分割的困擾,且全域式的表達方法不需事先定義行為的種類與持續時間,有利於後續開發之異常行為偵測與特定行為辨識視訊系統。 在異常行為偵測方面,收集日常生活之各種正常行為模式之後,設計相關行為表達之幾何特徵,並透過自動分群機制進行正常行為之群集訓練,以作為異常行為之偵測標準。在特定行為辨識方面,以獨立成份分析為基礎,自動擷取行為特徵並以特徵作為行為辨識之基準,同時本研究也提出了以獨立成份重建技術,量測特定行為之相似度,使得特定行為辨識之方法可進一步用於復健/運動評估等應用。 實驗結果顯示本論文所提出的智慧型視訊系統應用在安全監控、居家照護、醫療看護、人機互動、視訊影片檢索及復健/運動評估等應用,皆能夠有效偵測及辨識感興趣之行為,輔助管理人員進行有效監控以及降低人力及資源成本。

並列摘要


Intelligent video system (IVS) uses computer vision algorithms to automatically detect unpredictable abnormal events or identify specific activities in video sequences. IVS generally involves the following three steps: 1) low-level detection of foreground/moving objects from the background with a still camera, 2) spatiotemporal representation of motions in an image sequence, 3) high-level decision making for classification/recognition. This research concentrates on the development of intelligent video systems, including foreground segmentation, spatiotemporal representation, abnormal event detection and specific activity recognition. A mode-based background subtraction model for moving object detection in an image sequence is first proposed. It detects the most frequently occurring gray level of each pixel, instead of the mean, in the image sequence. It can quickly respond to changes in illumination and more accurately extract the silhouette of foreground objects against a low-contrast background. For spatiotemporal representation of a motion, this research uses a fast macro-observation representation that records the time-space change of motions of all moving objects in a scene without segmenting individual object parts. The proposed representation of motions can simultaneously represent both spatial context and temporal context of a motion, and is used for abnormal event detection and specific activity recognition. An abnormal event detection scheme is developed to detect unexpected events such as burglary and fighting in daily life. The discriminative features are extracted from the spatiotemporal representation of motions. A clustering mechanism is proposed to group similar activities sampled from image sequences of normal daily life. Any events deviating from the training ones are then classified as abnormalities. It allows fast computation of similarity measure for each new scene image. For specific activity recognition, an independent component analysis (ICA) based scheme is proposed. It uses the spatiotemporal representation of specific activities for training. The motion features of the spatiotemporal representation are automatically extracted by ICA basis image reconstruction. The proposed method can well perform for activity recognition in a disturbed background. It can be potentially applied to video surveillance, human-computer interaction, video retrieval, and rehabilitation/exercise evaluation.

參考文獻


[30] C.-C. Chen, J.-W. Hsieh, Y.-T. Hsu, C.-Y. Huang, 2006, Segmentation of human body parts using deformable triangulation, in: Proc. International Conference on Pattern Recognition, 1, pp. 355-358.
[99] P. C. Hsiao, C. S. Chen, and L. W. Chang, Human action recognition using temporal-state shape contexts, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2008, pp. 1-4.
[9] T. Kanade, R. Collins, A. Lipton, P. Burt, and L. Wixson, 1998, Advances in cooperative milti-sensor video surveillance, in: Proc. DARPA Image Understanding Workshop, 1, pp. 3-24.
[80] J. Liu, S. Ali, and M. Shah, 2008, Recognizing Human Actions Using Multiple Features, in: Proc. IEEE Conference on Computer Vision and Pattern Recognition, pp. 1-8.
[48] A. A. Efros, A. C. Berg, G. Mori, J. Malik, 2003, Recognizing action at a distance, in: Proc. International Conference on Computer Vision, pp. 726-733.

被引用紀錄


吳雪菁(2015)。高感染風險經皮穿刺傷後對醫療人員生理與心理之影響〔博士論文,國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342%2fNTU.2015.02934
黃惠如(2014)。臺灣醫療工作者發生高感染風險經皮穿刺傷之現況分析〔碩士論文,國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342%2fNTU.2014.02981

延伸閱讀