Design of compressive imaging masks for human activity perception based on binary convolutional neural network

Abstract

Many applications demand proper design and implementation of 0-1 binary compressive sensing (CS) measurement matrices. This paper presents a construction method for such binary CS measurement matrices by training a convolutional neural network (CNN) with 0-1 weights. The desired CS performance of resultant binary measurement matrices can be achieved by designing a proper CNN training procedure. For human activity recognition applications, such a sensing system is implemented with a small number of optical sensors and optical masks, which can achieve a high recognition capability with a far smaller amount of data than traditional cameras. In the experiments, the compressive sensory readings are classified using a basic K-Nearest Neighbor (KNN) algorithm to demonstrate the high sampling efficiency of hardware without compromising much the recognition performance.