Modern automated object detection systems are key tools in surveillance applications. These systems rely on computationally expensive computer vision algorithms that perform object detection on visual data created by surveillance cameras. Due to the nature of surveillance systems, this visual data must be processed accurately and in real-time. However, many of the frames that are created by the surveillance cameras may be of low importance, providing no useful information to the object detection system. Sub-sampling the surveillance data by prioritizing important camera frames can greatly reduce unnecessary computation. Several works have been conducted on dynamic visual data sub-sampling using various modalities of information (ie. spatial or temporal information) for prioritization. However, few works have used different modalities of information together for visual data prioritization. Furthermore, given the fast pace of the research space, only a small subset of works have implemented visual data prioritization with modern computer vision algorithms in mind. This work evaluates several individual prioritization metrics, that use different modalities of information to prioritize visual data, for use with modern object detection algorithms. This thesis presents an ensemble method that uses a KNN regressor to combine the best of the previously evaluated metrics. This dynamic approach was shown to increase the detection rate in an indoor surveillance scenario by over 60% compared to a static sub-sampling baseline.
Zircon - This is a contributing Drupal Theme Design by WeebPal.
