Pedestrian Recognition with False Positive Detection by Model-based Tracking

R. Miyamoto, H. Sugano, and Y. Nakamura (Japan)


Pattern Recognition, Pedestrian Recognition, Particle Fil ter, Stochastic Tracking, False Positive Detection


Nowadays, pedestrian recognition based on image process ing is widely tackled. Generally, pedestrian recognition is constructed by combining detection and tracking of pedes trians. However, accuracy of pedestrian recognition de grades since non-pedestrian objects are tracked once they are falsely detected as pedestrians. To overcome this prob lem, a novel pedestrian recognition by combining detection based on boosting and skeleton-based stochastic tracking with false positive detection is proposed. In the proposed scheme, false positives are detected based on the variance of predicted skeleton in a tracking phase. The experimen tal results by applying the proposed scheme to a sequence provided by PETS show that false positives can be detected by the proposed scheme based on the variance.

Important Links:

Go Back