Derin Öǧrenme ile Termal Kameralarda Görüş Açisi Kestirimi

In this study, a deep learning-based method is proposed to automate the field of view of the camera measurement process, which is applied to each camera coming out of the mass production line and currently performed with the human eye. This proposed method is fundamentally based on detecting the time instants when a certain target enters and exits the scene seen by the camera. To be detected these moments, improved YOLOv3 with ResNet50 hybrid architecture is proposed. It is aimed to create a target detection model with less training data. With this model, the camera field of view is estimated. By using two different test systems and two different thermal cameras, a dataset is created to be used in the training processes of the proposed hybrid architecture, Faster-RCNN, and YOLOv3 target detection architectures. In the performed camera field of view estimation tests using this dataset, it is seen that the proposed hybrid architecture has higher accuracy than other target detection architectures.