Title:  Content-adaptive Cross-layer Optimized Video Processing using Real-time Feature Feedback

Committee: 

Dr. Chatterjee, Advisor        

Dr. Romberg, Chair

Dr. Raychowdhury

Abstract:

The objective of the proposed research is to design a low-power video processing system capable of minimizing power consumption through graceful reduction of the quality of the processed signal.  Methods for post-processing signal analysis are proposed for determining the presence and quality of features essential to the high-level application goals.  Two applications, video encoding and object tracking, are selected for study based on their high demand on mobile platforms.  Methods for scaling the complexity of the entire system are proposed by simultaneously scaling down the input signal and algorithms to focus computational effort on information salient to the application.  Dynamic voltage and frequency scaling (DVFS) is proposed along side error detection techniques.  The voltage and frequency are to be scaled as low as possible (beyond design thresholds) without violating constraints on the quality of the application-specific measured features.  Finally, a cross-layer control system is proposed for determining the optimal complexity scaling and DVFS.  The control system will be charged with adjusting scaling levels to minimize power consumption while narrowly achieving the high-level application goals.