AI based smart surveillance

Project Title

AI based smart surveillance

Project Area of Specialization Artificial IntelligenceProject Summary

To make Pakistan secure we are trying to make such a system which is efficient to give the alert in case of human detection, intrusion detection, any anomaly event detections (for domestic and industrial use, for commercial use other features are used like car detection and emotion detection ). For this purpose, we use the raspberry pi as a processor and  NCS2 from fast inference with raspberry pi. The deep learning approach is used for this purpose. 

Project Objectives

The task of the human operator is not limited to watching videos and react to abnormal situations. After training and acquiring experience, a human operator is able to incorporate the area context in order to judge the events on the monitor. However, humans do make mistakes. Even more, it appears that they make a significant number of mistakes when they are watching surveillance monitors. The main reason is the nature of the task: passively watching multiple monitor screens where nothing special happens for a long period of time. In order to solve this problem, we use artificial intelligence for the analysis of the video and observe and note any unusual activity and annotate the video. The main points which we included in the analysis of the video are Human Detection, face detection, Camera Tamper Detection, Video Search/Summary, Object Classification, and mainly the abnormal activity detection which will be notified through email or via message. Further, our system is identifying the vehicle and estimating its speed through artificial intelligence. Our aim is to Incorporate Pakistan into the technology era.

 

Project Implementation Method

Methodology

In our system of AI-Based smart surveillance system, coming towards surficial idea of methodology, webcam is used to fetch the frames and forward that frame to raspberry pi 3B+ [RPI], RPI reprocess the frames and forward to Neural Compute Stick2 [NCS2], NCS2 then implement the algorithm and send the results back to RPI.

Coming towards the deep analysis about implementing this project. Numbers of the features that will be part of this system are human detection, object detection, anomaly detection, intrusion detection, restricted zone notifier, age, gender and emotion detection along with vehicle detection also measuring the speed of the vehicle.

Every first step in implement is training the models, GPUs are used for training purposes, which is the most time and capital take. After training of models, we have to move towards conversion, for this purpose NCS2 is used. NCS2 is Intel’s product for the purpose of inference of machine learning algorithms irrespective of which frame models are trained. At the core of this strategy is that the Myriad Vision Processing Unit (VPU), an AI-optimized chip for accelerating vision computing supported convolutional neural networks (CNN). According to Intel, Myriad VPUs have dedicated architecture for high-quality image processing, computer vision, and deep neural networks, making them suitable to drive the demanding mixture of vision-centric tasks in modern smart devices.

It is powered by Myriad X VPU that comes with 16 Shave Cores. Intel claims that it's a minimum of eight times faster compared to the previous version.

The OpenVINO™ toolkit quickly deploys applications and solutions that emulate human vision. Based on Convolutional Neural Networks (CNN), the toolkit extends computer vision (CV) workloads across Intel® hardware, maximizing performance.

OpenVINO™ can take models built with multiple different frameworks, like TensorFlow or Caffe, and use its Model Optimizer to optimize for inference. This optimized model can then be used with the Inference Engine, which helps speed inference on the related hardware. It also features a big variety of Pre-Trained Models already put through Model Optimizer.

By optimizing for model speed and size, OpenVINO enables running at the edge. This doesn't mean a rise in inference accuracy - this must be wiped out training beforehand.

After the training, optimization inference will be done and the result will be generated depending upon which model is used. After the generation of desired results now it’s time to make it user-friendly for that purpose we are focused on developing an App so that user can have access to all streams, able to watch them and will also get notifications of different events that will happen either abnormal event is captured or some is trying to enter the restricted zone then desired security actions will be performed by human or user.

Benefits of the Project

As a world is moving toward AI every security system is now working on the basis of AI while Pakistan still lacks in the state of art. So we are trying to offer such a system which enables Pakistan to deploy this state of the art in safe cities, traffic police, industries. So that Pakistan also enlists in countries which are better in development and safe in security. It will provide  Domestic Security Targeted Marketing and Traffic surveillance. 

Technical Details of Final Deliverable

1. In our Project as the name infers we are using artificial intelligence to achieve our goal of presenting a smart surveillance. For this purpose the most challenging feature which we have faced is training of models as in Pakistan there are only a few Universities which facilitate the GPU system for training for this purpose we are in need of buying the GPU system for training purpose.

2. Second Challenge which we have face is the increased FPS (frame per second) for better video streaming without any hanging issue. For this purpose, by doing a lot of research analysis we have used Intel neural compute stick 2 for this purpose for better fps.

3. Third, we have used Openvino  for the deep learning model conversion so we can use the model as they are compatible for stick only by the conversion

4. Different algorithm testing to achieve the best outcome for each feature. (for this purpose different literature review has done and implemented their methodology and compare)

5. Anomaly activity detection is not achieved on live stream till now so we are developing our own algorithm and also doing data set collection till now we have gathered a 100 GB data for training and now in need of testing our algorithm.

(NOTE)

The most important point is as the world is moving toward Artificial Intelligence we are in need to move toward AI as Pakistan still lack a lot in this field because of lack of facility as we face the same issues of the training as the code idea models are ready but the unavailability of GPU cause us some major technical problems

Final Deliverable of the Project HW/SW integrated systemCore Industry SecurityOther Industries Others Core Technology Artificial Intelligence(AI)Other Technologies OthersSustainable Development Goals Industry, Innovation and Infrastructure, Sustainable Cities and CommunitiesRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	70000
Rasbpeery pi kit	Equipment	1	13000	13000
Neural Compute Stick 2	Equipment	1	19000	19000
Graphical Processing Unit	Equipment	1	38000	38000