Academia.eduAcademia.edu

Figure 2 – uploaded by Selliishwarya S

See full PDFdownloadDownload figure
Database 9300 manually contour labelled pedestrian samples from 6389 images with separate training and testing sets TABLE I A few landmark research on direction of motion estimation (contd. Constraints, Limitations and Assumptions

Table 2 Database 9300 manually contour labelled pedestrian samples from 6389 images with separate training and testing sets TABLE I A few landmark research on direction of motion estimation (contd. Constraints, Limitations and Assumptions

Related Figures (18)

Fig. 1. 8 equiangular discrete directions with respect to field camera
Fig. 1. 8 equiangular discrete directions with respect to field camera
Fig. 2. Plots representing unique pattern of change in dimensions as the pedestrian moves in different directions
Fig. 2. Plots representing unique pattern of change in dimensions as the pedestrian moves in different directions
bounding box along with the displacement direction of its centroid. Thus, the aggregated feature is formed as a 1D array which is linear concatenation of temporal pattern of change in height, padding (for displacement direction of centroid) and temporal pattern of change in width. ‘ig. 4. Genesis of the aggregated feature vector
bounding box along with the displacement direction of its centroid. Thus, the aggregated feature is formed as a 1D array which is linear concatenation of temporal pattern of change in height, padding (for displacement direction of centroid) and temporal pattern of change in width. ‘ig. 4. Genesis of the aggregated feature vector
Fig. 5. Illustration for finding pd in different cases With a trade-off between the two factors outlined above, the value of n has empirically been found to be 5. This possibly happens due to the presence of partial occlusion or segmentation error (where height or width is not visible) persisting through consecutive 4 frames. Through this understanding, the size of padding can be modelled based on the nature of the database (and the efficiency of background subtraction applied on it) on which the method is supposed to work.
Fig. 5. Illustration for finding pd in different cases With a trade-off between the two factors outlined above, the value of n has empirically been found to be 5. This possibly happens due to the presence of partial occlusion or segmentation error (where height or width is not visible) persisting through consecutive 4 frames. Through this understanding, the size of padding can be modelled based on the nature of the database (and the efficiency of background subtraction applied on it) on which the method is supposed to work.
Fig. 6. Selection of pd reflecting classification through displacement of centroid Thus it can be well justified why an aggregate feature vector with padding (Case I shown above) will bear better directional
Fig. 6. Selection of pd reflecting classification through displacement of centroid Thus it can be well justified why an aggregate feature vector with padding (Case I shown above) will bear better directional
Fig. 7. State transition diagram of the HMM of proposed method to 50-60 iterations in case of 4 hidden states. This empirically justifies the selection of number of hidden states.
Fig. 7. State transition diagram of the HMM of proposed method to 50-60 iterations in case of 4 hidden states. This empirically justifies the selection of number of hidden states.
Fig. 8. Block diagram of the proposed method
Fig. 8. Block diagram of the proposed method
CASIA Dataset A CASIA Dataset B- NITR conscious walk database
CASIA Dataset A CASIA Dataset B- NITR conscious walk database
Fig. 10. Log Likelihood graph presenting machine learning of 6 discrete directions using HMM over CASIA A Datase
Fig. 10. Log Likelihood graph presenting machine learning of 6 discrete directions using HMM over CASIA A Datase
Fig. 11. Log Likelihood graph presenting machine learning of 5 discrete directions using HMM over CASIA B Dataset
Fig. 11. Log Likelihood graph presenting machine learning of 5 discrete directions using HMM over CASIA B Dataset
Fig. 12. Log Likelihood graph presenting machine learning of 8 discrete directions using HMM over NITR Conscious Walk Dataset
Fig. 12. Log Likelihood graph presenting machine learning of 8 discrete directions using HMM over NITR Conscious Walk Dataset
QUANTITATIVE COMPARISON OF DIRECTION ESTIMATION RESULTS OVER DIFFERENT DATABASE TABLE XI -3536 (c) 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. | http://www. ieee.org/publications_standards/publications/rights/index.html for more information.
QUANTITATIVE COMPARISON OF DIRECTION ESTIMATION RESULTS OVER DIFFERENT DATABASE TABLE XI -3536 (c) 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. | http://www. ieee.org/publications_standards/publications/rights/index.html for more information.
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved