• Vinay Nagar
• Vidit Gupta
• B. Baranidharan

Action recognition in videos that are taken from normal cameras has become crucial for various applications such as surveillance and action determination. Videos can contain some camera shake, obstructions, and noisy background. In normal action recognition method, a sizable number of confined motion features like space-time interest points (STIP) and motion scale-invariant feature transform (MoSIFT) are derived from videos. Then, all obtained features are calculated into a histogram vector using bag-of-words representation and then the vector-based classifiers are employed to do recognition in the test dataset. But noises and unrelated information creeps into the bag-of-words during the extraction and calculation of the obtained local features. In recent work, there is some method which takes leverage of labelled video but it is very difficult to collect a large number of labelled video for the training of the system. So, we are proposing a system which utilizes both labelled and unlabelled videos for learning and then will be able to learn more accurate classifier. We will first take the video and divide it into a number of different frames. The Laplacian smoothing transform (LST) feature is applied which will then identify the edges and then smooth the image using the Gaussian smoothing filter to reduce its sensitivity to noise.

LST, STIP, MoSIFT

"Action Recognition Using Mined Hierarchical Compound Features", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.6, Issue 3, page no.371-373, March-2019, Available :http://www.jetir.org/papers/JETIR1903F55.pdf

"Action Recognition Using Mined Hierarchical Compound Features", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.6, Issue 3, page no. pp371-373, March-2019, Available at : http://www.jetir.org/papers/JETIR1903F55.pdf