Performance evaluation of Multiple Target Tracking in the absence of reference data

Abstract

Evaluating the performance of Multiple Target Tracking algorithms is a crucial requirement for the design and validation of different applications. Most of the available evaluation metrics require knowledge about the true state of the estimated situation. However, in most practical applications, such reference data are not available. Thus, a system of performance evaluation metrics which do not require reference data is proposed in this paper. The presented measures evaluate the detection performance, accuracy, and quality of the system output. The approach is evaluated based on simulated data and demonstrated on the example of an automotive tracking application.