Human Task Recognition using CNN

Abstract

In this fast pacing world, computers are also getting better in terms of their performance and speed. It is capable of solving very complex problems like understanding an image, understanding videos and live capturing and processing of data. Due to advancement in technologies like computer vision, machine learning techniques, deep learning methods, artificial intelligence, etc., various models are being made so that prediction of outputs is made simpler and of high accuracy and precision. Our project model is built using a convolutional neural network (CNN). Our dataset consists of 599 videos in which 100 videos was assigned to each category of basic human actions like Running, Boxing, walking etc. In this project, we have used a set of labelled videos which was used to train our three models. The CNN is used as a base network in all the three models to process all the videos from the dataset, that is, to read all the frames and convert into heat maps. Out of our three models, the best model is used to give prediction for the actions performed in the video. The results show that with better algorithm techniques, the performance of the model is also improved.