Detection of Overlapping and Small-Scale Targets in Indoor Environment Based on Improved Faster-RCNN

This paper proposes an improved Faster-RCNN network to detect small-scale and overlap targets in indoor environment. In the improved Faster-RCNN model, ResNet-50 was used to replace VGG16 as the backbone network to extract multi-layer features of images, and then shallow feature images with rich detailed information were fused with the deep feature images with abstract information. In addition, Soft-NMS is used to replace NMS to improve the performance of overlapping target detection. In self-made dataset, the mAP of the improved model was improved by 1.45% relative to the unimproved model.