Learning invariance from transformation sequences
Neural Computation
Optimal, unsupervised learning in invariant object recognition
Neural Computation
Neural Networks for Pattern Recognition
Neural Networks for Pattern Recognition
Autonomous Driving Goes Downtown
IEEE Intelligent Systems
Fast object recognition in noisy images using simulated annealing
ICCV '95 Proceedings of the Fifth International Conference on Computer Vision
A Pattern Classification Approach to Dynamical Object Detection
ICCV '99 Proceedings of the International Conference on Computer Vision-Volume 2 - Volume 2
Motion-Based Recognition of Pedestrians
ICPR '98 Proceedings of the 14th International Conference on Pattern Recognition-Volume 2 - Volume 2
ACCV'10 Proceedings of the 2010 international conference on Computer vision - Volume part II
KES'05 Proceedings of the 9th international conference on Knowledge-Based Intelligent Information and Engineering Systems - Volume Part IV
Semi-supervised training set adaption to unknown countries for traffic sign classifiers
PSL'11 Proceedings of the First IAPR TC3 conference on Partially Supervised Learning
| Hi-index | 0.10 |
This contribution describes a classification module integrated into an image processing system consisting of a subsequent detection, tracking, and classification stage that extends its "knowledge" about a certain object class by autonomous in situ training. In our scenario, a classifier initially trained with front and rear views of pedestrians mainly in dark clothes subsequently extends its recognition capabilities in a first step towards lateral views of pedestrians in dark clothes, in a second step towards lateral views of pedestrians wearing both dark and bright clothes. Although supervised training algorithms are applied, at no point during the autonomous in situ training processes is an interaction with a human operator necessary; all training labels are autonomously generated by computing track-specific class assignments from time-step-specific class assignments. It is demonstrated that a significant improvement of the recognition performance with respect to new appearances, i.e., lateral views, of pedestrians, can be achieved without "forgetting" the initial front and rear views.