TY - GEN
T1 - A method for evaluating performance of joint tracking and classification
AU - Zhang, Le
AU - Lan, Jian
AU - Li, X. Rong
N1 - Publisher Copyright:
© 2015 IEEE.
PY - 2015/9/14
Y1 - 2015/9/14
N2 - Joint tracking and classification (JTC) is rapidly gaining momentum recently. Algorithms have been proposed for this problem. However, performance of tracking and classification has been evaluated separately without considering their interdependence. In this paper, we propose a joint measure, named joint probability score (JPS), to account for tracking error, misclassification and their interdependence. The basic idea of JPS is to measure the difference between the cumulative distribution functions (CDFs) of the ideal JTC and the one to be evaluated. Moreover, performance of tracking and classification can be unified by applying CDF. The proposed method is unit free and positive definite. Also, it has close connections with stochastic dominance and the so-called continuous ranked probability score. Two examples illustrating our JPS are presented. The results demonstrate that JPS reflects well the joint performance of tracking and classification.
AB - Joint tracking and classification (JTC) is rapidly gaining momentum recently. Algorithms have been proposed for this problem. However, performance of tracking and classification has been evaluated separately without considering their interdependence. In this paper, we propose a joint measure, named joint probability score (JPS), to account for tracking error, misclassification and their interdependence. The basic idea of JPS is to measure the difference between the cumulative distribution functions (CDFs) of the ideal JTC and the one to be evaluated. Moreover, performance of tracking and classification can be unified by applying CDF. The proposed method is unit free and positive definite. Also, it has close connections with stochastic dominance and the so-called continuous ranked probability score. Two examples illustrating our JPS are presented. The results demonstrate that JPS reflects well the joint performance of tracking and classification.
UR - https://www.scopus.com/pages/publications/84955090581
M3 - 会议稿件
AN - SCOPUS:84955090581
T3 - 2015 18th International Conference on Information Fusion, Fusion 2015
SP - 499
EP - 506
BT - 2015 18th International Conference on Information Fusion, Fusion 2015
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 18th International Conference on Information Fusion, Fusion 2015
Y2 - 6 July 2015 through 9 July 2015
ER -