Video related to moving object detection, shadow removal, tracking, and behavior Recognition

The graduation design of graduate students focuses on sports target detection, tracking, shadow removal, behavior analysis, etc. But where does the video come from?

There are many online websites:

Moving Target Detection: http://www.research.ibm.com/peoplevision/performanceevaluation.html

Videos such as petsd2tec1 are often used here. In the four outdoor videos, the moving targets of two videos are blocked by a big tree, which can be used for target tracking.

In the indoor video, there are 1 to 3 people, and multiple people can follow the target.

1. Outdoor sequences (4 videos from pet2001 with ground truth)

Petsd1tec1. Avi
Petsd1tec1. pvann

Petsd1tec2. Avi
Petsd1tec2. pvann

Petsd2tec1. Avi
Petsd2tec1. pvann

Petsd2tec2. Avi
Petsd2tec2. pvann

2. Indoor sequences (11 videos with ground truth)

Indoorgttest1.avi
Indoorgttest1.pvann

Indoorgttest2.avi
Indoorgttest2.pvann

Threeperson_circles_comp_0_quad0.avi
Threeperson_circles_comp_0_quad0.pvann

Threeperson_circles_comp_0_quad2.avi
Threeperson_circles_comp_0_quad2.pvann

Threeperson_circles_comp_0_quad3.avi
Threeperson_circles_comp_0_quad3.pvann

Threeperson_together_split_comp_0_quad0.avi
Threeperson_together_split_comp_0_quad0.pvann

Threeperson_together_split_comp_0_quad2.avi
Threeperson_together_split_comp_0_quad2.pvann

Threeperson_together_split_comp_0_quad3.avi
Threeperson_together_split_comp_0_quad3.pvann

Twoperson_line_circle_comp_0_quad0.avi
Twoperson_line_circle_comp_0_quad0.pvann

Twoperson_line_circle_comp_0_quad2.avi
Twoperson_line_circle_comp_0_quad2.pvann

Twoperson_line_circle_comp_0_quad3.avi
Twoperson_line_circle_comp_0_quad3.pvann

 

In the bottom other areas of research, you can see other IBM research results in video processing. However, it is a pity that there are only result demos and no related papers can be referenced.

Shadow Detection aspect: http://cvrr.ucsd.edu/aton/shadow/

There is a video of two highways, a video of a parking lot, and an indoor video. There is also a split frame of one image. However, I think that the highway video speed is too fast, which often leads to invalid motion target detection algorithms. Let alone shadow. The parking lot is good. It's quite useful. It is not easy to use in the room, because the clothes and backgrounds worn by people are too similar.

 

However, it is difficult to find videos on abnormal behavior recognition. The biological recognition and Security Technology Research Center of the institute of automation of the Chinese Emy of Sciences has established a complete database:

Http://www.cbsr.ia.ac.cn/china/demos?20ch.asp, the content inside is very good. There are also presentations of research results.

The only drawback is that there is no way to directly download the video. He can only mail it. It is probably afraid that the research results will be stolen by hackers! In fact, it is not expensive. videos do not cost money, as long as postage. You can take a closer look at the instructions on the website.

Several databases have been discovered recently:

Cvonline databases:

Http://homepages.inf.ed.ac.uk/cgi/rbf/CVONLINE/entries.pl? Tag363

Korean intelligent Media Lab Database: http://imlab.postech.ac.kr/

University of Dublin face database: http://dsp.ucd.ie /~ Prag/

Video Sequences of American Sign Language (ASL ):

Http://csr.bu.edu/asl/html/sequences.html

MIT computer science and Artificial Intelligence Lab database:

Http://db.csail.mit.edu/

Let's take a look at the specific content.

 

I tried shooting by myself, but it was very difficult to shooting:

First, if you want to perform behavior detection, you must take close-to-scene videos. Remember that there must be no reflective things on the background (such as glass or tile), especially in the light, they often detect motion targets.

Second, pay attention to the differences between your clothes and the background (or even shadow. Otherwise, it is difficult to distinguish.

Again, if the camera is shot in close-range mode, try to move it in parallel to the lens direction. Otherwise, some parameters are useful in some places and cannot be used in other places.

No activity videos can be downloaded for free. If anyone knows, please leave a message below. Thank you very much.