Development of intelligent techniques to detect generic movable objects in video

Abstract

Terrorist acts and crime rate have increased the demand for improved automated video surveillance system. The main goal of this research work is to design generic movable object detection and tracking effectively. To attain this goal, an automated integrated movable object detection and tracking is designed with three phases. An integrated system is developed in the first phase to track and classify multiple moving objects. At first, the curvelet transform, soft thresholding and frame differencing is used to localize the moving object. Then, texture, color and shape features are extracted from the localized object through applying the feature extraction techniques. Essentially, Speeded Up Robust Features (SURF) is employed for extracting the shape feature; the Enhanced Local Vector Pattern (ELVP) for extracting the texture features; and Histogram of Oriented Gradients (HOG) for extracting the color features. Subsequently, the Genetic Algorithm (GA) is imposed to reduce these extracted features and fused together to form a single vector. Finally, these fused reduced features are fed into the Extreme Learning Machine (ELM) to classify multiple moving objects. This newly proposed integrated system achieved good performance in classification/detection of multiple moving objects compared to other existing schemes under a variety of scenarios revealing different environmental conditions and camera orientations. newline