Activity Recognition and Motion Characterization

+ MultiClass Object classification in  Video Surveillance systems  There is growing demand for automated public safety systems for detecting unauthorized vehicle parking, intrusions, un-intended baggage, etc. One impacting factor for these applications is object detection and recognition in surveillance systems. This is challengeable problem since the purpose of the surveillance videos is to capture wide landscape of the scene; resulting in small, low resolution and occluded images for the objects. The goal of this project is to design and implement recognition system for objects in outdoor surveillance videos. In this paper we extracted as many as 25000 features to make an inherent study of the system with many parameter settings. We managed to build up an efficient Object classification system with different configurations. The system was evaluated based on various parameters and models mainly SVM and AdaBoost. Configuration of the systems includes both domain shift features and feature selection.

system Architecture
online Run of the system