Hello, welcome to our IOT group 45 I uh security camera project. I am here to explain to you a bit of how our code works and how it all comes together. So the title of our project is object detection Security Camera and the the domain this falls into is security. So the IOT we developed is uh we use the Raspberry Pi Secure uh the camera module and we did take inspiration from surveillance cameras used by governments and companies for houses and street lamps. Everything else uh a camera works by when it sees an object, it tries to figure out if it is a person or not. Originally, this code was able to detect any objects. But due to the Raspberry Pi um limitations and obviously overheating and stuff like that. I had to limit it to people. So if it is a person, then it does draw an outline around the object and also displays a percentage that, that it thinks. So if it is above a certain percentage or a certain threshold, it will detect that it is a person. Usually the value we set in here is um 45 to 60% above. So we use Python with the computer vision library to achieve this, it streams the camera output to a website for which we used flask. Our motive was to create a security camera that displays its output somewhere and we use the web because it would be easy to implement and wouldn't require us to design any physical equipment. And on the whole, our product fulfills the role of security camera as it outputs itself to the website and informs the user for the texts a person. So um this project does contain a library of names for previous, the previous functionality of detecting any object. This was taken from a tutorial online on youtube. And if I were to scroll down this whole code, this is the whole entire code for our project. It does take in imports such as uh open CV flask for our flask server and numpy um under coco dot names is where all the, the sorts of names it would that it would take from if it were to identify one of these objects. So our code code is here, we did struggle at first to implement a way of uploading this to a web server. Um Due to our Raspberry Pi's version, I'll be able to install a couple applications that would be needed. So we simply just stuck with the FL server. Um a local host which um the person using the IOT, our IOT would simply just type in local host on their web browser um colon 8000 and they would see the broadcast of the Raspberry Pi. So talking about products um on the market today that people could buy that are similar to ours. And there are quite a few which are similar to the security camera, which we developed a notable example is NES CM IQ INDO, which is similar to our product. The Nes CM also has an advanced computer vision capability and they use A I algorithms in order to detect and identify people and then also has the ability to send an alert to user smartphone when a person is detected in front of the camera, which our IOT does isn't able to do. The person would act, would have to go on to the website and see for themselves. Uh It will outline the shape of the detective person and will then also provide a confidence score which through the help of computer version determines the likelihood of accuracy of the detection. There is the option for a continuous stream to the app which allows for remote viewing of a user's home by the given links on their website. It takes you there. And uh another honorable mention is the Arlo Ultra two, which is a wireless security camera system with advanced motion detection features and four KHDR video. And it may not provide a clear outline of items it has identified, but it does provide customizable motion zones and notifications for certain activities such as human detection. It also sends live video to the Arlo app so they can be viewed and monitored remotely. So to conclude it does there, there are a couple of IOT projects and um products out there that are quite similar to ours. So the value proposition of our I our IOT system is that our IOT security camera systems value proposition. And primarily it's seamless integration of cutting edge technology which improves its convenience as well as safety. Um by leveraging computer vision algorithms, our camera can accurately detect and identify people in its field of view. Um But the issue is with object detection and in our case, um human detection is that the frames on the camera are quite low. We were not able to get a smooth 30 fps or even 60. Um Unfortunately, the Raspberry Pi your Hes up too much and there's no other way apart from passive cooling which we were limited to. So, but the object detection and human detection works absolutely fine. We just have to limit it to people. It, this does detect multiple people at once. If it is shown in the frame. If the, the the people are in the front of the camera, this can be limited to only one person. But um that might not give us the best values as passing people or most people in a view might not give, you know, the person a clear idea of how many people are there or how many people have been identified. So as I was talking earlier, um the fact that the device can determine the percentage accuracy of its reading is significantly a significant benefit for users as it allows them to the peace of mind in regards to how sure they can be of of the reading because the reading is shown to the user as well on the outline that it provides when the camera detects a person, it is streamed in real time to a website for which we did use fast as I mentioned previously, which allows users to take swift action whether it involves checking the live feed to assess the simulation or notifying authorities if necessary, the capacity capacity to remotely monitor one's property facilitates proactive security ma management and offers peace of mind. So this code does not have much to it. It is a single file which just takes in libraries from other places. Uh We do have a demo web stream pie which we did take from a support website explaining on local host flash servers, but this isn't currently in use. It was just for us to get an idea on how to implement into our object identifying Python file. And the whole code is completely um it doesn't, it's not, it doesn't get affected by any of the web stream pie. Uh All we have to do is transfer elements that would allow us to stream onto a logo host and um provide us with the feed that we need. We also needed a index template for this. And um our index template is completely is simple. It just tells us um I don't know why that is talk detection, but it should say object action system. And um oh any sorry. And it just takes in the video feed that we that is within object identify pi and uh transmits it to our Python server um pre previously for that dark detection. Um We did try a open source file given by a youtuber that um provided a tutorial on how to detect animals, but primarily dogs. But that wasn't needed for our project. We wanted to try and get humans as it was the whole point. And with this code, it, it works flawlessly just apart from the frames, we are able to change the size and input scale of the, the window. But for what we have already, it's perfectly fine. It's perfectly visible for all users. And yeah, there is not much to it. Thank you very much for listening.