Pedestrian Detection Inspired by Appearance Constancy and Shape Symmetry

Abstract

The discrimination and simplicity of features are very important for effective and efficient pedestrian detection. However, most state-of-the-art methods are unable to achieve good tradeoff between accuracy and efficiency. Inspired by some simple inherent attributes of pedestrians (i.e., appearance constancy and shape symmetry), we propose two new types of non-neighboring features (NNF): side-inner difference features (SIDF) and symmetrical similarity features (SSF). SIDF can characterize the difference between the background and pedestrian and the difference between the pedestrian contour and its inner part. SSF can capture the symmetrical similarity of pedestrian shape. However, it's difficult for neighboring features to have such above characterization abilities. Finally, we propose to combine both non-neighboring and neighboring features for pedestrian detection. It's found that nonneighboring features can further decrease the average miss rate by 4.44%. Experimental results on INRIA and Caltech pedestrian datasets demonstrate the effectiveness and efficiency of the proposed method. Compared to the state-ofthe-art methods without using CNN, our method achieves the best detection performance on Caltech, outperforming the second best method (i.e., Checkboards) by 1.63%.

Soft-Cascade Learning with Explicit Computation Time Considerations

“…It is necessary to remark that the two set of evaluations have been made in different computer configurations, which can be seen in the differences between their and ours ACF detector results. We can observe that our detectors are comparable with the results reported by Cao et al [5]. Crosstalk detector has better computation time of 45.40 frames per second followed by our ACF+GLS with 31.86 fps and ACF with a mean of 31.305 fps.…”

“…We observe that ACF+GLS processes 1.15 more frames per second than the original ACF cascade. We compare our results with the one reported by Cao et al [5] in Figure 7. It is necessary to remark that the two set of evaluations have been made in different computer configurations, which can be seen in the differences between their and ours ACF detector results.…”

“…One more time, we observe that ACF DISC+CLUS+GLS process 3.03 more frames per second that the ACF DISC cascade. We compare our results with the ones reported by Cao et al [5] and Zhang et al [33] in Figure 11. Let us highlight that these evaluations were conducted in more powerful computer architectures than our, some of them (CCF, CompACT-Deep and RPN+BF [33]) even using GPU architecture.…”

“…Nevertheless, as the number of constraints (6) and (7) depends on the product of the cascade size L and the training set size N , it cannot be applied directly for large scale soft-cascades as the model becomes too huge to be solved in finite time even using the best outstanding commercial BIP solvers. Hereafter, an iterative search procedure that allows overcoming the previous drawbacks is detailed.…”

“…Crosstalk detector has better computation time of 14.10 frames per second followed by our ACF DISC+CLUS+GLS with 9.88 fps and ACF with a mean of 9.49 fps. The detectors with the best compromise [5] and [33].…”