Novel Metrics for Performance Evaluation of Object Detection Algorithms

This paper proposes novel metrics to evaluate the performance of object detection algorithms in video sequences. The proposed metrics allow to characterize the methods being used and classify the types of errors into region splitting, merging or merge-split, detection failures and false alarms. This methodology is applied to characterize the performance of five segmentation algorithms. These tests are performed in the context of object detection in outdoor scenes with a fixed camera.