A Performance Evaluation of Object Detections by Progressive Quality Improvement Approach

Due to the widespread of recent object detection technologies, various real-time object detection systems such as human detection systems and car detection systems are deployed in banks, airports and so on. In most of these systems, camera devices continuously send their original recorded images to processing computers even when their target objects for detection are not recorded. This causes a large amount of communication traffic such as exceeds bandwidth usage, delays for data transmission. The communication traffic can be reduced by continuously sending rough images and re-sending clear images only when clear images that have objects recorded are required for the applications. For instance, a image that recorded a human. However, detecting objects in the rough images decreases its accuracy. In this paper, to evaluate the performance of our proposed progressive quality improvement approach, we investigate the accuracy of object detections changing the qualities of images.