2. Envejecimiento cognitivo
2.3. Cambios afectivos en el procesamiento de la información
2.3.1. Sesgo positivo Teoría de la Selectividad Socio-emocional
In the following section, prototype T is evaluated to determine how well the tracking phase performs as a hand-tracking framework.
6.6.1.1 Prototype T
In Figure 6.9 a graphical representation of the results of the tracking success rates of prototype T for each video of a signed gesture performed by the participants is shown (see Appendix D, TableD.2 for the detailed results).
Table 6.3: Summary of the average hand-tracking success rates of prototype T for each signed gesture in both environments.
Single hand Both hands with-
out occlusion
Both hands with occlusion
Success rate
Sign 3 Sign 11 Result ≥ 80%
Sign 1, 4, 5, 6, 7, 8, 10
Sign 14, 15, 19 70% ≤ Result < 80%
Sign 2, 9 Sign 13, 16, 17, 18 Sign 21, 22, 23, 24,
25, 26, 27, 29
60% ≤ Result < 70%
Sign 12, 20 Sign 28, 30 Result < 60%
The results in Figure6.9represent the average hand-tracking success rate of each signed gesture in both environments. In Table 6.3 the results indicate that 12 signs obtained an average success rate greater than 70% with sign 3 and sign 11 obtaining success rates greater than 80%. These results were the set of signs that involved a single hand and both hands without occlusion. In addition, 14 signs obtained an average success rate of between 60% and 70%, with six of those signs obtaining an average success rate very close to 70%. These results were the set of signs that involved both hands with and without occlusion, in which more than 50% of these signs involved occlusion. Therefore 60% of the signs obtained an average success rate close to or more than 70%. Furthermore, only four signs obtained an average success rate below 60%.
Since prototype T forms the foundation of the hand-tracking framework, the results presented were very encouraging. Overall, the comparison between the different signed gestures indicates that tracking two hands while distinguishing between them increases the complexity of hand-tracking. Furthermore, it also indicates that occlusion between the hands affects hand-tracking and is one of the main reasons for hand-tracking failure.
Chapt er 6. E xp er im ental R es u lts a nd A na ly si s 109
Figure 6.9: The average hand-tracking success rates of prototype T for each signed gesture in both environments.
Chapter 6. Experimental Results and Analysis 110 In addition to these factors, the assumption in prototype T that the object being tracked is a hand could be a factor resulting in lower results. This assumption is investigated in the following section.
When comparing how well the hands were tracked in the different environments us- ing prototype T (as seen in Figure 6.10), the results indicate that 17 signs obtained a higher result in constrained environments when compared to unconstrained environ- ments, which suggests that the background in unconstrained environments negatively affects hand-tracking. The results also indicate that 70% of the signs using a single hand were higher in a constrained environment than an unconstrained environment in terms of the average success rate, while half of the signs using both hands were the same in either environment in terms of average tracking success rate (refer to Table D.2 in Ap- pendix D). This signifies that the movement of two hands is more likely to be affected by moving objects in the background when compared to single hand movements. In terms of accuracy when considering the signing of the participants, for more than half of the participants the prototype’s average tracking success rate was greater than 70% (see FigureD.1in AppendixD) and for four of the participants the average success rate of prototype T was greater than 80%. Furthermore, four participants obtained an average success rate of between 60% and 70%, and five participants obtained an average tracking success rate of below 60%. Overall, the results show an equal variation between the different body types and skin-colour tones. The hand-tracking results can therefore not be attributed to the participants that performed the gestures, but rather to other factors such as the moving objects in the background, light inconsistencies in a frame and the complexities of gestures using both hands.
When comparing related systems (with respect to hand-tracking) to prototype T, the comparison is restricted to passive approaches that use 2D video input streams. It is also restricted to those that followed an appearance-based method. Many of these related systems have opted for a subjective evaluation; however, those that have pursued an objective evaluation have evaluated their systems using their personal datasets.
In comparison to the single hand-tracking systems (in simple environments) of Rautaray and Agrawal [128], Coogan et al. [36] and Fogelton [54], prototype T provides a more accurate approach. It achieves above 70% for 80% of the total number of signs using a single hand and is able to handle occlusion between the face and hands. Even though
Chapt er 6. E xp er im ental R es u lts a nd A na ly si s 111
Figure 6.10: A comparison of the average hand-tracking success rates of prototype T for each signed gesture in constrained and unconstrained environments.
Chapter 6. Experimental Results and Analysis 112 Liu et al. [97] improved the flocks-of-features algorithm of Fogelton [54] in terms of dis- tractors in the background, their hand-tracking is still negatively affected when moving the tracked hand in front of the face, whereas this scenario is successfully handled using prototype T.
When comparing prototype T to the particle-filter frameworks of Liu and Zhang [98], Spruyt et al. [144], Ongkittikul et al. [120] and Chen et al. [32], it shows a comparable performance in terms of tracking both hands. In addition, it is able to distinguish between the right and left hand when compared to Liu and Zhang [98] and Ongkittikul et al. [120]. Moreover, prototype T extended the algorithm of Argyros and Lourakis [10], but with several improvements: it is able to distinguish between the right and left hand, it does not regard the hand as a different object when it recovers from tracking failure, and it is able to track both hands in unconstrained environments.