Evaluation of Human Detection Algorithms in Image Sequences - Advanced Concepts for Intelligent Vision Systems - page 128

Table 1. Performances evaluation of the different interpretation algorithms developped
within the CAPTHOM project

Set 1 Set 2 Set 3 Set 4
Viola [8], sf =1.1 0.614 0.672 0.565 0.597
Viola [8], sf =1.4 0.719 0.707 0.105 0.436
Viola [8], sf =1.5 0.758 0.739 0.092 0.451
Viola [8]+BS, sf =1.1 0.429 0.642 0.050 0.276
Viola [8]+BS, sf =1.4

0.618 0.747 0.071 0.380

Viola [8]+BS, sf =1.5 0.663 0.745 0.082 0.405

CAPTHOM 0.338 0.089 0.043 0.176

5 Conclusion and Perspectives

We presented in this paper the potential interest of a global evaluation met-
ric for the development of industrial understanding algorithms. The originality
of the proposed measure lies in its ability to simultaneously take into account
localization and recognition aspects together with the presence of over- or under-
detection. Concerning the foreseen application, industrial partners involved in
the project also have in mind to extend the system for car park video surveillance.
In that case, the detection and distinction between different classes could be in-
teresting and give even more sense to the misclassifcation error introduced in the
evaluation metric. We are actually working on the use of taxonomy information
for ponderating the misclassification error. The introduction of a distance matrix
between classes taking into account their more or less important similarity could
improve the adaptability of the proposed metric. For some applications, some
misclassifications could have less repercussions than others. As an example, it
could be suitable to less penalize an interpretation result where a bus is recog-
nized as a truck, as these two objects are very similar, than an interpretation
result where a bus is recognized as a building.

References

1. Cucchiara, R., Grana, C., Piccardi, M., Prati, A.: Detecting moving objects, ghosts,
and shadows in video streams. IEEE Transactions on Pattern Analysis and Machine
Intelligence (PAMI) 25(10), 1337-1342 (2003)
2. Deselaers, T., Keysers, D., Ney, H.: Improving a discriminative approach to object
recognition using image patches. In: Kropatsch, W.G., Sablatnig, R., Hanbury, A.
(eds.) DAGM 2005. LNCS, vol. 3663, pp. 326-333. Springer, Heidelberg (2005)
3. Dalal, N., Triggs, B.: Histograms of oriented gradients for human detection. In:
Proc. IEEE Computer Society Conference on Computer Vision and Pattern Recog-
nition (CVPR), pp. 886-893 (2005)
4. Everingham, M., Van Gool, L., Williams, C.K.I., Winn, J., Zisserman, A.: The
PASCAL Visual Object Classes Challenge 2008 (VOC 2008) Results,
http://www.pascalnetwork.org/challenges/VOC/voc2008/workshop/index.html