Table 4.

Detailed results obtained with the proposed method and an alternative association strategy that assigns parts to one another by minimizing Euclidean distance. The table includes true positives (TP), false positives (FP), false negatives (FN), precision, recall, and F-measure for different part detection thresholds.

Part Detection Threshold	Vector Matching						Euclidean Matching
	TP	FP	FN	Recall	Precision	F-Measure	TP	FP	FN	Recall	Precision	F-Measure
0.10	20,217	27	525	0.975	0.999	0.987	19,170	1181	1572	0.924	0.942	0.933
0.15	20,160	20	582	0.972	0.999	0.985	19,127	1156	1615	0.922	0.943	0.932
0.20	20,092	17	650	0.969	0.999	0.984	19,058	1154	1684	0.919	0.943	0.931
0.25	19,999	13	743	0.964	0.999	0.981	18,971	1141	1771	0.915	0.943	0.929
0.30	19,865	10	877	0.958	0.999	0.978	18,851	1135	1891	0.909	0.943	0.926
0.35	19,675	7	1067	0.949	1.000	0.973	18,675	1162	2067	0.900	0.941	0.920
0.40	19,413	3	1329	0.936	1.000	0.967	18,418	1190	2324	0.888	0.939	0.913
0.45	19,029	2	1713	0.917	1.000	0.957	18,077	1195	2665	0.872	0.938	0.904
0.50	18,408	2	2334	0.887	1.000	0.940	17,526	1269	3216	0.845	0.932	0.887
0.55	17,287	2	3455	0.833	1.000	0.909	16,568	1281	4174	0.799	0.928	0.859
0.60	15,227	2	5515	0.734	1.000	0.847	14,871	1294	5871	0.717	0.920	0.806
0.65	11,929	0	8813	0.575	1.000	0.730	12,184	1189	8558	0.587	0.911	0.714
0.70	7565	0	13,177	0.365	1.000	0.534	8543	860	12,199	0.412	0.909	0.567
0.75	3261	0	17,481	0.157	1.000	0.272	4523	430	16,219	0.218	0.913	0.352
0.80	692	0	20,050	0.033	1.000	0.065	1414	90	19,328	0.068	0.940	0.127
0.85	53	0	20,689	0.003	1.000	0.005	143	3	20,599	0.007	0.979	0.014
0.90	1	0	20,741	0.000	1.000	0.000	5	0	20,737	0.000	1.000	0.000