162 A.R. Ortiz et al. P.‚jD; Mj/ /P.Dj‚; Mj/P.‚jMj/ (20.5) where P.‚jD; Mj/ is the posterior probability density function (PDF) of the parameters ‚, for model Mj, given the observation D. P.‚jMj/ is the prior PDF of the parameters ‚and it represents the knowledge of the parameters before updating. P.Dj‚; Mj/ is the likelihood of the occurrence of the measurement Dgiven the vector of parameters ‚andmodel Mj. Mj is the model presented in Sect. 20.2.1 In this research, samples of the posterior are obtained to derive statistics of the parameters. The samples are generated using the Markov chain Monte Carlo (MCMC) methodology [18–20]. The MCMC is a derivation of the Monte carlo sampling algorithm, where the samples distribution is based on an equilibrium condition. The Markov chain algorithm defines the probability of the next step based on the probability of the current step. Further details about model updating of human-structure interaction models using this approach can be found in Ortiz 2016 [21]. 20.3 Experimental Setup The identification of the trajectory of the pedestrian’s CoM involves the use of an infrared video camera. In order to track the trajectory, the artificial vision system needs to move along the distance that the pedestrian walks, keeping an optimal distance, otherwise images will not have the same resolution. Therefore, the experimental measurement of the human body is performed using a mobile platform Microsoft Kinect V2 for Windows and a Laptop, as seen in Fig. 20.1. The mobile platform is located in front of the human body. Acquisition of the depth images of the Kinect sensor is performed once the mobile platform is moving at constant velocity. The relative velocity of the human to the mobile platform is measured using depth images. The velocity of the mobile platform is measured using two integrated quadrature encoder which provide a resolution of 64 counts per revolution of the motor shaft. This resolution corresponds to 4480 counts per revolution of the gearbox’s output shaft. Therefore, synchronization of the velocity signals is required to calculate the actual velocity of the pedestrian. Figure 20.2 shows a picture of the test. The camera and the computer are on the mobile platform ahead of the pedestrian. The mobile platform is on rails in order to reduce the noise generated by surface irregularities. The acquisition and processing of depth images is performed using own developed algorithms in Matlab [22]. The developed algorithm performs a calibration of the depth images in order to obtain a rectified point cloud of the scene in front of the Kinect V2. The scene acquired on the depth images is complex since it includes all the surroundings. The surroundings are removed using a spatial threshold volume. As a result of the inclination angle of the Kinect V2 depth camera, the orientation of the segmented human body needs to be corrected. The ground floor surface is also identified (assumed constant) and the angle ˛ is obtained. This surface is used for correction of the inclination angle. A rigid transformation of Fig. 20.1 Picture of the mobile platform carrying the computer and the MS Kinect sensor (at the top)
RkJQdWJsaXNoZXIy MTMzNzEzMQ==