Bayesian Approach for Detection Classification

Abstract

The objective of this paper is to develop and test a software system that uses incomplete information from a collection of sensors to classify different objects present in a particular area with a pre-specified probability. The objects in the study are referred to as vehicles called the Bus and the Truck. Intruding vehicles move across a designated geographical area. Sensors that have been placed in that area detect vehicles and calculate probabilities that a vehicle is of a specific type conditioned on the type of vehicle that is actually detected. The goal is to determine unconditional probabilities that a given detection is of a particular type. The main idea is to find which vehicle is located at a geographical point in a designated area using the Bayesian approach to calculate the probabilities for this detection classification problem. Each sensor tries to detect the vehicle based on its sensing radius, which is nothing but the distance between the sensor and the vehicle. To test the methodology, I assumed that the probabilities vary depending on the color of the vehicles. For example, if a vehicle is red in color, it is assumed to be easier for the sensors to classify than if it is blue. The framework uses Bayesian inference to calculate the probabilities and to distinguish two types of moving vehicles. Experiments are conducted to find the number of sensors that successfully distinguish two types of moving objects with a given probability threshold. In the future the Detection Classification Model can be used to distinguish any number of objects with the mobile sensors and also some obstacles included in a designated geographical area.