peddetection
所属分类:模式识别(视觉/语音等)
开发工具:matlab
文件大小:5382KB
下载次数:278
上传日期:2012-04-08 21:06:06
上 传 者:
圆溜溜
说明: 行人检测的源程序,基于HOG算法,能实现对行人的检测。
(The source of the pedestrian detection, pedestrian detection, based on the HOG algorithm.)
文件列表:
ped-demo-fast-8x8x1.1\approx_models.mat (5156076, 2009-06-16)
ped-demo-fast-8x8x1.1\compute_features.m (729, 2009-03-27)
ped-demo-fast-8x8x1.1\compute_features_fast.m (4984, 2009-10-23)
ped-demo-fast-8x8x1.1\compute_features_fast.m~ (4979, 2009-03-27)
ped-demo-fast-8x8x1.1\compute_features_scale_space.m (1894, 2009-03-27)
ped-demo-fast-8x8x1.1\compute_feature_dim.m (165, 2009-03-27)
ped-demo-fast-8x8x1.1\compute_gradient.m (1585, 2009-03-27)
ped-demo-fast-8x8x1.1\compute_gradient_features.m (1168, 2009-03-27)
ped-demo-fast-8x8x1.1\concat_features.m (453, 2009-03-27)
ped-demo-fast-8x8x1.1\cumsum2D.m (215, 2009-03-27)
ped-demo-fast-8x8x1.1\det.png (15457, 2009-03-27)
ped-demo-fast-8x8x1.1\draw_det.m (579, 2009-03-27)
ped-demo-fast-8x8x1.1\get_sampling_grid.m (404, 2009-03-27)
ped-demo-fast-8x8x1.1\mex_feature.cc (7126, 2009-03-27)
ped-demo-fast-8x8x1.1\mex_feature.mexa64 (12280, 2009-03-27)
ped-demo-fast-8x8x1.1\non_max_sp.m (1986, 2009-03-27)
ped-demo-fast-8x8x1.1\normalize_response.m (608, 2009-03-27)
ped-demo-fast-8x8x1.1\ped_demo.m (558, 2009-06-16)
ped-demo-fast-8x8x1.1\pos_test.png (324416, 2009-03-27)
ped-demo-fast-8x8x1.1\run_detector.m (3758, 2009-06-16)
ped-demo-fast-8x8x1.1\test_feat.m (418, 2009-03-27)
ped-demo-fast-8x8x1.1 (0, 2012-04-08)
An Implementation of a pedestrian detector based on multiscale HOG like features
and pyramid match kernel based SVM. The classifier based on the paper:
----------
Classification using Intersection Kernel SVMs is Efficient,
Subhransu Maji, Alexander C. Berg, Jitendra Malik
CVPR 2008, Anchorage, Alaska.
---------
Run the ped_demo.m to run the pedestrian detector. It loads the precomputed
models and runs the detector on a test image (from the INRIA person dataset).
you can change the scaleratio to make the code go faster.
It is currently set at 2^(1/8) and it takes about 6.7s to run the detector
over 9046 windows and classify them. The classification time is 0.***s.
The folder contains (approximate) pretrained models computed from the
INRIA training data. If you want to change the feature parameters then you
have to retrain models. We currenly use LIBSVM to train the models and
use the precomp_models (inside libsvm-mat-2.84-1-fast.v3) to compute the
approximations. The classification is done using fiksvm_predict.
Add libsvm-mat-2.84-1-fast.v3 to the path. This is the code for running
fast prediction using the ideas from our CVPR'08 paper. The approximate
models use 300 bins in each dimension for approximation.
--
Subhransu Maji
www.cs.berkeley.edu/~smaji
近期下载者:
相关文件:
收藏者: