ADDRESSING THREE PROBLEMS IN EMBEDDED SYSTEMS VIA COMPRESSIVE SENSING BASED METHODS

摘要

Compressive sensing is a mathematical theory concerning exact/approximate recovery ofsparse/compressible vectors using the minimum number of measurements called projections.Its theory covers topics such as l1 optimisation, dimensionality reduction, informationpreserving projection matrices, random projection matrices and others. In this thesiswe extend and use the theory of compressive sensing to address the challenges of limitedcomputation power and energy supply in embedded systems. Three different problemsare addressed. The first problem is to improve the efficiency of data gathering in wirelesssensor networks. Many wireless sensor networks exhibit heterogeneity because of the environment.We leverage this heterogeneity and extend the theory of compressive sensingto cover non-uniform sampling to derive a new data collection protocol. We show that thisprotocol can realise a more accurate temporal-spatial profile for a given level of energyconsumption. The second problem is to realise realtime background subtraction in embedded cameras. Background subtraction algorithms are normally computationally expensivebecause they use complex models to deal with subtle changes in background. Thereforeexisting background subtraction algorithms cannot provide realtime performance on embeddedcameras which have limited processing power. By leveraging information preservingprojection matrices, we derive a new background subtraction algorithm which is4.6 times faster and more accurate than existing methods. We demonstrate that our backgroundsubtraction algorithm can realise realtime background subtraction and tracking inan embedded camera network. The third problem is to enable efficient and accurate facerecognition on smartphones. The state-of-the-art face recognition algorithm is inspiredby compressive sensing and is based on l1 optimisation. It also uses random projectionmatrices for dimensionality reduction. A key problem of using random projection matricesis that they give highly variable recognition accuracy. We propose an algorithm tooptimise projection matrix to remove this performance variability. This means we can use fewer projections to achieve the same accuracy. This translates to a smaller l1 optimisationproblem and reduces the computation time needed on smartphones, which have limited computation power. We demonstrate the performance of our proposed method onsmartphones.