Vibhor Agarwal

This research project was a part of India Innovation Challenge Design Contest’ 2018 organised by DST, Govt. of India in collaboration with IIM Bangalore. Our team was selected out of 10146 teams for the contest. It aims at pro-actively detecting suspicious activities like molestation against women and automates the current manual Surveillance process to enhance the security standards for women in the society. • Uses hybrid neural networks containing stacked Convolutional Neural Networks and Long Short Term Memory Networks sequentially to capture spatial and temporal information from the live video streams. • Hardware used is Beaglebone black as a micro-processor along with CCTV cameras to process frames and predict suspicious activities in real-time.

Implemented a sequential Encoder-Decoder Long Short Term Memory Networks based model to translate the input text in German to its corresponding English text. Used Bilingual Evaluation Understudy (BLEU) score as an evaluation measure against the reference sentences in the test set.