Vision based Human Activity Recognition

Automated systems (particularly surveillance systems) are in demand for the classification of various human actions as the number of cameras grows day by day. Furthermore, human action recognition (HAR) has emerged as one of the most appealing study subjects across a wide range of computer vision ap

Project Title

Vision based Human Activity Recognition

Project Area of Specialization

Artificial Intelligence

Project Summary

Automated systems (particularly surveillance systems) are in demand for the classification of various human actions as the number of cameras grows day by day. Furthermore, human action recognition (HAR) has emerged as one of the most appealing study subjects across a wide range of computer vision applications. The precise detection of human activity on an uncertain basis, on the other hand, remains a mystery. There has been limited study on human action recognition processes in real situations, which drives us to pursue research in this application space. In this paper, we examined several approaches, including CNN, CNN+LSTM, MobileNet, and Inception-V3. Experiments are carried out on the UCF-101 dataset to demonstrate the effectiveness of new models.

Project Objectives

The goal of a HAR is identifying actions or activities done by a person or a group of people. HAR has been a research interest for various groups over the past years and it is considered to be an active research area due to applications such as content-based video analysis and retrieval, visual surveillance, HCI, education, medical as well as abnormal activity recognition, etc. [26]. Activity recognition is essential to humanity, since it records people’s behaviors with data that allows computing systems to monitor, analyze, and assist their daily life [27]. There are two mainstreams of HAR systems: video-based systems and sensor-based systems. Sensorbased systems utilize on-body or ambient sensors to dead-reckon people’s motion details or log their activity tracks. While Video-based systems use cameras to take images or videos to recognize people’s behaviors [1]. Due to an increase in the usage of cameras, automated systems are in demand for the classification of such activities using computationally intelligent techniques. HAR’s framework can be classified into two methods: Action Detection and Action Classification (which is also divided into two sub methods: Action Representation Method and Interaction Representation Method). HAR is further divided in In video based HAR, a general steps to complete the process includes: a) Action representation – which consists of feature extraction and encoding, b) Dimensionality reduction techniques – original features are transformed by removing the redundant information through different models such as: PCA, RBD, LDA, and KDA etc., and c) action analysis-based HAR – action classification techniques are used which mainly include traditional ML as well as DL techniques like CNN, RNN, LSTM, GRU and GAN etc. [26].

The objective of our FYP project was to perform research on recognizing different human activities from videos or still images. For improving existing result for HAR, we have reviewed existing HAR approaches form past research papers. So, that we could figure out more efficient hybrid model in terms of good accuracy in performance and results compare to previous found approaches and models.

Project Implementation Method

We used different deep learning models like as CNN, Inception-V3, and Mobile-Net over UCF101 to see which models will work best for HAR. However, the individual model testing results were not as good as planned. As a result, we attempted a hybrid technique, such as CNN + LSTM and Inception-V3 + LSTM, to improve the accuracy of our testing results.

Technical Details:

Models Implemented: CNN, LSTM, Inception-v3, MobileNet, Inception+CNN, Inception+LSTM, CNN +LSTM

Dataset: UCF101

AI libraries: Tensor Flow, OpenCV, 

Tool/Compiler: Anaconda Jupter Nodebook 

GPU: Navida GPU 2070 Super

Benefits of the Project

The scope of HAR, also motivated us to research Human Activity Recognition on videos because the success of research will lead us to cover more kind of different activities. Most Importantly, this research will open the new doors for research in Video Capturing.

Furthemore, our research could help in varies area such as servillance system, patient/employee/childer/customer monitering, 

Technical Details of Final Deliverable

FYP Report + Research paper + IPYNB (Jupyter Notebook files of Implemented Models)

Final Deliverable of the Project

Software System

Core Industry

Education

Other Industries

Others , Security

Core Technology

Artificial Intelligence(AI)

Other Technologies

Big Data

Sustainable Development Goals

Industry, Innovation and Infrastructure, Sustainable Cities and Communities

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com