Human Action Recognition Based on Tracking Features - Foundations on Natural and Artificial Computation - page 475

Fig. 2. (a) Evolution in time of the bounding box. (b) Variation of Lx and Ly associated

to the bounding box.

4 Experimental Results

Core TM 2Duowith3GB
RAM. The parameter configuration selected for the results achieved is summa-
rized in Table 2.

All the experiments were performed on a 3GHz Intel R

Table 2. Parameter configuration

Parameter description value

N

Number of particles 30

I

Number of iterations in the MA stage 5

[ th R ,th G ,th B ] [45 , 45 , 45]

Measurement model thresholds

[ min F ,max F ]

Diffusion range for [ x,y ]

[ − 5 , 5]

in the system model and
in the mutation method

[ min G ,max G ] Diffusion range for [ x, y ] [ − 1 , 1]

in the system model

[ Δx o ,Δy o ,ΔLx o ,ΔLy o ] Movement length values [2 , 2 , 1 , 1]

in the improvement method

C

Penalty parameter for 32768

the SVM classifier

γ

0 . 00012207031 Kernel parameter for SVM