Information-Based Bit Allocation for Cooperative Visual Sensing in Vehicular Networks

Cooperative visual sensing among multiple vehicles has raised increasing concern with the development of connected vehicular systems. However, it remains challenging for effective cooperative sensing under communication constraints due to the lack of theoretic basis and efficient design for communication resource allocation in visual sensing networks. This paper establishes an information-based bit allocation framework for cooperative visual sensing in vehicular networks, where visual observations captured by on-board cameras are quantized before transmitted to the fusion center for 3D environment model estimation. Our aim is to allocate the quantization bits among observations in the network to minimize the overall estimation error. Specifically, we first design a Fisher information matrix (FIM)-based metric to evaluate the cooperative visual sensing error, and characterize the impact of quantization on sensing with geometric interpretations based on the relation among the 3D physical space, the quantization modulation space and the information eigenspace. Then we present closed-form investigations for the optimal bit allocation among observations in the network, and derive its dependence on network topology and communication constraints. Moreover, we propose a hierarchical bit allocation scheme and a convex relaxation-based scheme for cooperative visual sensing, and compare their performance with the generalized Breiman, Friedman, Olshen, and Stone (GBFOS) algorithm and the uniform allocation scheme. Numerical results validate that our schemes achieve near-optimal sensing performance under communication constraints with a significant reduction of computation load.