OpenNPD

Project of object detection.

Usage:

1. Add needed librarys to 3rdpart folder(opencv).

2. Make: cd src; make clean; make RELEASE=1; cd ..;

3. Detect example: sh ./script/test_detect_demo.sh

4. Training example: Please refer to ./script/npd_train_demo.sh

Note:

1. Thereis a npddetect::prescandetect function for faster detection with some lose on recall. Additional parameter stepR refers to pre-scan step size compared to the original scan step size ( float more than 1 ) . The thresR refers to the threshold to reject the window( float in [0-1] ).

Result:

• ROC:

  

• Speed:

| image size | window size | cores | time (ms) | | :---: | :---: | :---: | :---: | :---: | | 640x480 | 20x20 | 1 | ~50 |

• ROC for npddetect::prescandetect:

• Speed for npddetect::prescandetect:

params	image size	window size	time(ms)
none	1920x1080	20x20	532.400239
stepR = 2, thresR = 0.2	1920x1080	20x20	344.154205
stepR = 2, thresR = 0.3	1920x1080	20x20	282.128798
stepR = 3, thresR = 0.2	1920x1080	20x20	286.230415
stepR = 3, thresR = 0.3	1920x1080	20x20	226.091203
stepR = 4, thresR = 0.3	1920x1080	20x20	202.147923

References:

Project page

@article{

 Author = {Shengcai Liao, Member, IEEE, Anil K. Jain, Fellow, IEEE, and Stan Z. Li, Fellow, IEEE},

 Title = {A Fast and Accurate Unconstrained Face Detector},

 Year = {2014}

}