Introduction to Video Analytics

A free video tutorial from EDUCBA Bridging the Gap
Learn real world skills online
Rating: 4.3 out of 5Instructor rating
330 courses
414,988 students
Introduction to Video Analytics

Learn more from the full course

Video Analytics using OpenCV and Python Shells

Through this training we shall understand and learn how to perform video analytics with OpenCV.

02:12:14 of on-demand video • Updated December 2018

Understand and learn how to perform video analysis with OpenCV.
Learn Color Models, Image Loading and Image Thresholding
Master Open CV and Object Detection
English [Auto]
Hello guys so they're going to the local switches. Video analytics. No you know you very well know about the scores because today we do know that X is very important in almost many or all all of the feeds. If you see in our market then in every business we do it on the text comes first because in every field we require surveillance that the security and the United text is very helpful in that field. OK so Will it come to mean definitions or what is that. So see they can say that this food is for the process of analyzing videos. So basically in this we analyze videos and that's the name is VDO and takes all they can say we gather intelligence from them or they can say we are extracting some meaningful information from the dealers. Now right digital videos because we are converting those videos in digital form and then they're extracting information from them. So this is what we do analytics similarly as opposed to video compression but attempts to exploit the redundancy in digital video or forward using sys and other Dix's concerned with understanding the content of the you know little what is video compression. So in fact what we do is we just reduce the size of our VDU we remove that tendency present in that video and we just attempt to reduce that size. But what is analytics and other pixels just opposite of that that does in other text not opposer it can say but it is concerned with understanding the content of video. What we do is actually have actually Von Boosey Oregon's expecting some meaningful information from that video content. Is that what that is the deal in politics OK. Now see if I can use this video analytics it is used for research in computer vision. All they can say pattern analysis machine intelligence surveillance can every day. And transportation video content analysis or Ragan's intelligent feed. So these are some things where this video out of Texas important. OK so that that's even better better them and in discourse basically that will focus on like the result in a computer vision. So basically in this field and the scores Vivan focus on computer vision topics like object detection motion detection object tracking motion like these topics with overcovered underscores. Okay. And so that also focus on like that done analysis Sutherland's system like that. OK. So it uses computer vision of the God of them that enable it to perceive or see and machine that is to interpret learn and draw inferences. Right. So in this field of ill use computer vision Ill go to them. So in our next video to show you some computer vision that go to them and begin to work on them to get some meaningful output. So they're very useful to Hill and the goal of what is the goal of that it will be seen understanding the understand that scene. And that's very different from motion detection so if I can make IMC it has a motion detection because we are just getting the information from that VDU. So what is that that it's getting extracting information from that content. So this is your view in politics and motion detection is part of that. In addition to detecting motion it qualified the motion as an object. Understand the context around the object and is able to direct the object to the scene. So let's say of an feel like motion detection dog. In addition to detecting motion work it will do is it understand the context our own dog. And it will drag the object to the scene. So that does the complete. DANKOSKY analytics. OK. Now see if we go to it there are basically four letters of the two and I did X which is segmentation classification tracking and activity to cognition. OK so what is the first step that the segmentation saw segmentation. Is the process of detecting changes and extracting relevant changes for further analysis and qualification. So this process is the process of detecting changes. Right. And like Suppose pixels that have changed are the effortless foreground pixel and those that do not change are background fixed unordered means that's up to you. And this way can detect motion of any object in that video by just classifying that video in two types of pixel that is foreground pixel and background X a shorter background so those pixels which are constant throughout the video that is video not moving the turn or any motion in that and for grown folks that have motion. So by dividing by by doing that classification we can detect that motion. OK. So this process is not art that is segmentation classifying those pixels like that. There's a segmentation and detecting changes to them. OK. So there you see that desktop that can detail enough for those videos this is just normal definition of that and they can see the first layer of your widget analytics. The result of that is one or more foreground block a block or is that a collection of connected pixels. So block the separate or particular object from its background in a video or they can see it in a scene right now see next as your classification or what this classification This is their second layer. This is the process of qualifying each block and assigning a class labor to it. So whatever I do is after getting blocked and segmentation visually assigning them a class label. So this is overclassification. This will result in a broad categorization of each blob and be sufficiently distinct classes that are sports and vehicle and IIM-A etc. so bad that they can detect it. What does the type of that object like that object the object that is moving or that undergoing any motion is a person vehicle or animal like that. So this comes under classification process. Next as you're tracking so tracking of classified block block that isn't the segmentation unclassified block that goes in classification specking of them takes place over multiple flavors object move through the field of view. So it is a problem of blood Association for each blob and the starting thing the position of that blow needs to be identified because that position keeps on changing. And we have to take that position. So as far as tracking a tragic ligand indeed calculated for the object by connecting expositional multiple streams so basically what they're doing is they are just dividing our in multiple streams and rivers Volcan each and every frame of that video. And they track the motion or they can say the position of block. They just keep moving in each frame. And by this vacante trajectory and VIGEN track the object. OK. The last step is your activity recognition which is the final step. The combined result of classification tracking correlating the tracks of multiple blobs to and for the activity are getting in the way. So after knowing about the object Mosha its classification or its class labeler tracking. Now finally the vertical Mizer activity or they can say it's motion therefore it is actually doing. So this is what activated a cognition. For instance let's take an example of full blocks corresponding to people persuasively come closer. This can be interpreted as like conver they are converting something. OK. If Bill block Van Gogh responded to the call and another as a person happened to mode this could be interpreted as a person is getting into a vehicle. So this is vord is activity cognition so I'll those three Lilah's This comes and this will finally go Gisors or does the activity undergoing Andraka the deal. OK. So these were four years off the Alexandrov look on them and Beavan I have to show you these steps one by one and are next Tauriel. So there's probably going to be very interesting of course and you will be you'll enjoy very much discourse. So OK now see I show you one picture. And in this picture the people are moving like that and do they consider they can consider them as BLOBs because the remaining portion is constant that's not moving. So that can be considered as background grown people who are moving. They can be considered as blob. And we are in the first step whatever to do is to segment them and then in case of classification they recognize that they are people they will categorize them as people. So in the fosterling or whatever it is that they will just get there. The people at our blog recognize block and then in seconds Sturtevant Roadmaster there are people that didn't find the category of them just people and the target degree whatever it over correct the motion that they are moving they're crossing that zebra crossing line and the fourth Goudeket even recognize that they're like venturing here and there. So this is what this is your read an addiction. And finally analyzing that what is happening there or is undergoing in that real anger extracting that meaningful information from it. So this is for this is the do in politics. Next example is like that in we are detecting objects. So see here there are so many rectangles that detecting objects present and that. So like they're post-president you are like that. So this is what this is object detection. So the so this is your do not take scores. And in the next year tutorial they look on each load and see so many examples of it also see some interesting go to Adams in computer vision and in discourse so let's ready also for learning discourse. And let's see them let's see the next story.