Human Detection System using Different Depths of the Resnet-50 in Faster R-CNN

Locating people in an image or video sequence automatically is a challenging task due to different environmental conditions and wide variety of features. On the other hand, because of its variety of applications, automated human detection is a very popular area. This study proposes a deep learning model for automated human detection. In order to apply this task, the Pascal VOC database has been used for training and testing stages. The feature extraction step is performed using different depths of the Resnet-50 in a faster R-CNN model. Then, the obtained features are used for the classification step. When the results are compared for different depths of the Resnet-50, it is observed that the first 40 depth of the Resnet-50 showed the better performance than the first 46 depth. Also, the faster RCNN model used in this study outperforms the other studies used the same Pascal VOC database in the literature. The proposed system can detect humans indoor or outdoor environments. Also, it produces promising results in human detection task even in occlusion situations. It can be used for different applications such as monitoring people in public areas with security, health, or energy efficiency proposes, etc.