1. Yun-FuLiu Jing-MingGuo Che-HaoChang
Low resolution pedestrian detection using light robust features and hierarchical system
Yun-FuLiu
Jing-MingGuo
Che-HaoChang
Pattern Recognition,
Volume 47, Issue 4, April 2014, pp 1616–1625
Reporter: LIM ZI YI

2. Outline Introduction Problem descriptions
Proposed probability-based pedestrian mask pre-filtering
Orientation/magnitude-based AdaBoost algorithm
Experimental results
Conclusion
Personal Remark

3. Introduction
How to deal with the prospective complexities of background and light conditions is still a key issue in this field.

4. Introduction
Methods with temporal information consider movement or depth information have an inherent drawback that when the camera is moving or pedestrian is not standing still, the backgroundand foreground will be misclassified.
common and intuitive features, e.g., color and brightness, the performance can be easily affected by the dresses of pedestrians or environmental lighting conditions.

5. Introduction
This study presents a hybrid pedestrian detection scheme for significantly reducing detection time.
two topics are focused in this work:
(1) The feature descriptors for pedestrians, and
(2) the resolution of a pedestrian image.

6. Problem descriptions
Features for pedestrian
Sample resolution

7. Features for pedestrian
Fig. 2. Conceptual diagram of the edgelet feature applied to pedestrians.

8. Sample resolution
Another issue which directly affects driving safety is the practical detection distance.
Fig. 4. Relationship between the height of pedestrian and the distance from camera.

9. Sample resolution
Fig. 5. The difference between the same pedestrian captured with different resolutions.

10. Proposed probability-based pedestrian mask pre-filtering(PPMPF)
Initial weight mask generation
Refinement
Fig. 6. Algorithm of the proposed pedestrian detection scheme.

11. Initial weight mask generation
Two main components of the first stage:
probability-based pedestrian mask
the corresponding probability table (histogram table)
This study uses a convolutional pedestrian mask detection strategy

12. Initial weight mask generation

13. Initial weight mask generation
(4)
(5)

14. Initial weight mask generation
Fig. 7. Pedestrian pre-filtering by orientation mask.

15. Refinement

16. Refinement
Fig. 8. Pedestrian pre-filtering by magnitude mask.

17. Orientation/magnitude-based AdaBoost algorithm
Features
Training procedure
Detection process
Fig. 9. Flowchart of the proposed training process.

18. Features
(10)
(11)

19. Training procedure The AdaBoost can be divided into two parts:
voting.
the process trains the sorted datasets to obtain the classifiers, and uses the trained classifiers to construct a strong classifier to predict the samples of interest.

20. Training procedure
In this training procedure, the iteration updates the weights to upgrade the effect of the weak classifier to a strong classifier.
As the training process is completed, a voting mechanism is applied to form a strong classifier which can yield a precise detection capability.

21. Training procedure
Fig. 10. Orientation/magnitude-based features obtained from the average value of the blocks which are not easily affected during walking.

22. Detection process
Fig. 11. Flowchart of the orientation/magnitude-based detection process.

23. Experimental results Datasets: INRIA
Daimler Chrysler Pedestrian dataset
(15)
(16)

24. Experimental results
Fig. 12. Precision-recall plots with different features and the corresponding derived ways (1-D and the Sobel edge detector are considered).

25. Experimental results
Fig. 13. Precision-recall performances of the HOG, Haar-like, edgelet , and the proposed feature obtained with
the INRIA dataset and
the Daimler Chrysler Pedestrian dataset.

26. Experimental results
Fig. 14. Practical detection results with various features under diverse scenarios.

27. Proposed entire system (with PPMPF)
Experimental results
Table1 Detection speed comparison.
HOG
Haar-like features
Edgelet features
Proposed feature
Proposed entire system (with PPMPF)
Detect windows (s)
62.35
957.78

28. Conclusion
lighting change and low- resolution scenario, are considered simultaneously for pedestrian detection application.
the proposed scheme can offer an accuracy similar to that of the HOG, while the proposed scheme is much faster
the processing efficiency is similar to that of the edgelet, while the propose scheme is of higher accuracy.

29. Personal Remark On Table 1, Author didn’t explain what the unit is.
We can use other features like depth information to increase accuracy(but time of computing will become slower).
If need to apply such system on driving safety system, we can use laser radar to replace this type of method.

30. Thank you for your listening !

31. Q&A