Contextual learning in Video Analytics for Human pose Detection using Bayesian Learning and LSTM

Abstract

With the increase in the number of crimes in the city, we are in need of a Smart surveillance camera that detects anomalies in advance. In real-world object detection identity switching and object interactions are difficult and retain identities. Due to a lack of situational awareness real-time object detection and tracking lack semantic information. Surveillance cameras are installed everywhere, and we can’t identify peoples who might be a potential threat to security, Surveillance camera needs to be monitored all the time. Existing algorithm concentrate on feature aggregation at the pixel level. A novel method is proposed to track human different movements and positions encompassing deep and detailed features. The main goal of this paper is to propose a feature aggregation at a semantic level that will prevent threats in advance by introducing a deep learning technique with Contextual inference-based object detection using the Bayesian Rule which incorporates semantic relations between classes to recognize the location. It also integrates the relationship between the object in unseen classes which helps to identify located instances and predicts the location and extracts context features for superclass prediction.