在網路上的監控影像智慧分析需要較高識別度品質的影片，但是會顯著的增加網路頻寬要求。由於高動態物體影像在無線網路傳輸下造成影像品質下降問題，成為了智慧影像監控分析是否成功的關鍵。在本篇論文中，提出了一種基於物件特性的編碼方法，用以保持無線網路上影像資料流的穩定品質。影像品質與物件動態之間的反比關係（即，由於出現大而快速移動的物件而導致影像品質下降）在統計上可被表示為線性或非線性模型。本文針對線性模型提出了一種使用健全的M-estimator統計量的回歸演算法來建構針對不同bitrate的線性模型，應用線性模型來預測增強影像品質所需的bitrate增量。在非線性模型中採用二次多項式回歸演算法來建構針對不同bitrate的模型，利用迭代方法預測最適合的編碼bitrate。實驗中建立了一個模擬的無線網路環境，以驗證在不同無線網路情況下提出的方法，進行了具有各種物件動態的真實監控影像的實驗，以評估該方法的性能。實驗結果表明，相對於視覺和定量方面而言，以較低的無線網路頻?即可達到所需的串流影像品質。;An intelligent analysis of surveillance videos over networks requires high recognition accuracy by analyzing good-quality videos, which requires a significant amount of bandwidth. Degraded video quality due to high object dynamics during wireless video transmission causes more critical issues for smart video surveillance success. In this thesis, an object-based source coding method is proposed to maintain steady video streaming quality over wireless networks. The inverse relationship between video quality and object dynamics (i.e., decreasing video quality due to the presence of large and fast-moving objects) is statistically defined as a linear or nonlinear model. A regression algorithm based on robust M-estimator statistics is proposed to construct the linear model considering different bitrates. The linear model is used to predict bitrate increments required to improve video quality. A quadratic polynomial regression algorithm and an iterative method for predicting the most suitable encoding bitrates are used to develop the nonlinear model considering different bitrates. A simulated wireless environment is set up to verify the proposed method in various wireless scenarios. Experiments with real surveillance videos of a variety of object dynamics are conducted to evaluate the method’s performance. Experimental results show that the proposed method significantly improves video streaming in both visual and quantitative aspects when using lower wireless bandwidth.
