Accuracy Based Rewarding for Sensors in Noisy Collaborative Point Cloud Acquisition Environments

A common challenge in point-cloud acquisition environments, such as realizing traffic safety and autonomous driving, is to determine the placement of sensors and workers and the rewards to be paid to them. Game theory is available as an extremely powerful tool for the problem of determining appropriate rewards for deployed sensors and workers and several previous studies have proposed methods for determining rewards using game theory. However, these methods do not consider the affection of the recognition accuracy of downstream tasks by the AI to develop these appropriate rewards. In this paper, we propose a novel characteristic function of game theory by considering the recognition accuracy of AI. To define our function, we investigate how the number of viewpoints and the noise level of the point cloud affect the classification accuracy. In addition, we analyze which part of the point cloud the recognition model focuses on by using SHAP, a method to improve the explainability of machine learning based on the Shapley value.