Table 5.

Average recall, precision, and $F1$-score for each camera site used in the test dataset. The average was calculated based on five trained Faster R-CNN with ResNet50 models compared with five models trained with motion-enhanced images. The macro- and micro-average metrics cover results from all seven camera sites and weeks. The best macro- and micro-average of recall, precision, and $F1$-Score are marked with bold.

	FR-CNN	Motion	FR-CNN	Motion	FR-CNN	Motion
Camera	Recall	Recall	Precision	Precision	F1-Score	F1-Score
S1-0	0.051	0.262	0.032	0.758	0.037	0.385
S1-1	0.141	0.413	0.112	0.488	0.112	0.435
S2-0	0.305	0.529	0.250	0.650	0.274	0.576
S2-1	0.355	0.496	0.398	0.599	0.374	0.532
S3-0	0.404	0.487	0.538	0.840	0.459	0.612
S4-0	0.178	0.365	0.539	0.891	0.267	0.515
S4-1	0.496	0.585	0.262	0.634	0.337	0.603
Macro	0.276	0.448	0.305	0.694	0.266	0.522
Micro	0.300	0.446	0.344	0.751	0.320	0.555