Performance Evaluation for Multi-Camera Object Detection with Non-Overlapping Views

The proliferation of multi-camera systems in surveillance and monitoring applications demands robust methodologies for evaluating the performance of multiple object detection, especially in scenarios where camera fields of view do not overlap. This review paper systematically observes the key performance evaluation parameters pertinent to multiple object detection in multi-camera setups with non-overlapping fields of view. We explore traditional metrics such as precision, recall, and F1-score, as well as advanced metrics like Multiple Object Tracking Accuracy, Multiple Object Tracking Precision, and the recently proposed multi-view Higher Order Tracking Accuracy. Furthermore, we discuss the challenges unique to nonoverlapping FoV scenarios, including object re-identification, temporal synchronization, and spatial calibration. The paper also highlights existing datasets and benchmarks that facilitate the evaluation of such systems. Our comprehensive analysis aims to guide researchers and practitioners in selecting appropriate evaluation metrics and methodologies for developing and assessing multi-camera object detection systems with non-overlapping views.