Spatiotemporal Activity Mapping for Enhanced Multi-Object Detection with Reduced Resource Utilization

The accuracy of data captured by sensors highly impacts the performance a computer vision system. To derive accurate data, system must be capable identifying critical objects and activities in field reconfiguring configuration space real time. majority modern reconfiguration systems rely on complex computations thus consume lots resources. This may not problem for with continuous power supply, but it can major set-back employing limited Further, to develop an appropriate understanding scene, correlate past present events scene sensor’s view (FOV). address abovementioned problems, this article provides simple yet efficient framework reconfiguration. performs spatiotemporal evaluation generate adaptive activity maps, based which are reconfigured. maps contain normalized values assigned each pixel FOV, called sensitivity, represents impact or FOV. temporal relationship between is developed utilizing standard half-width Gaussian distribution. further proposes federated optical-flow-based filter determine Based re-configured align center most sensitive area (i.e., region importance) field. proposed tested multiple surveillance sports datasets outperforms contemporary terms multi-object tracking (MOTA).