Chapter 3 - Visual information processing and transmission in Wireless Multimedia Sensor Networks: a deep learning based practical approach

Abstract

The constant miniaturization in Complementary Metal Oxide Semiconductor (CMOS) based devices and the prolific advancement in cognitive algorithms has enabled the integration of varieties of heterogeneous sensing instruments on one sensing platform, the Wireless Multimedia Sensor (WMSens) node. As a result, a diverse set of multimodal information (image, video, temperature, humidity, acoustics, air quality, seismic activity, etc.) can be collected from the set of these miniature sensing instruments, which can be prioritized based on the level of importance. This chapter discusses one of the prominent modules in the WMSens, the vision system. Subsequently, we discuss the image processing and computer vision algorithms respectively in WMSens, the functional relationship between the visual processing algorithm and visual information transmission in WMSens and the Wireless Multimedia Sensor Networks (WMSN) as a whole, the applicability, compatibility, issues, and performance comparison of hand-crafted feature-based computer vision techniques and Deep Learning (DL) based computer vision techniques in WMSens. Finally, we provide a potential way forward to design the algorithms and protocols for the processing and transmission of visual information in WMSN and the related issues that remain unaddressed.