Application to Pedestrian Detection and Object Detection

Abstract

Automatic driving systems, object detection is essential. A logic of fusion presented combining the benefits of these two types detectors of objects, taking into account the properties of classical and deep learning techniques. Theoretically, a link between detection performance and detector type can be established. The numerical study to increase detection performance is based on the established theoretical relationship. In addition, an enhancement strategy is proposed that the designs of the sub-detectors are guided by this principle for improved overall performance. The utility of this combination methodology is illustrated in the identification of pedestrians using a trained by a machine on attribute the conventional detectors or human being. On the training datasets as well as additional different datasets to complete several comparative experiments using the classical and CNN detectors have been undertaken. It is a guarantee to improve detection performance and flexibility to different application settings with a simplified network.