DeepCamouflageAI: Deep Learning-Based Camouflage People Detection and Segmentation
Amna Arfa Rafique, Saeed Ur Rehman, Muhammad Kamran Imran, Khurram Shehzad, Hisham Yousaf · 2023
Camouflage hides information about any object in the background by efficiently merging foreground and background pixels. However, differentiating between these two-level pixels is a challenging task. Existing techniques work on camouflage object detection (COD), but camouflage people detection (CPD) needs to be explored more. Therefore, we have developed a CPD model for People detection and segmentation. The proposed model is based on deep convolutional neural network layers and transfer learning techniques. We exploited Mobile Netv2 as a base technique and presented a model that efficiently detects People from camouflage backgrounds. The model is trained and evaluated using the CAMO and COD10K standard datasets. Later evaluation is performed using the CPD dataset of 1K camouflaged people images. Comparative analysis between different techniques is performed and evaluated. The proposed model performed very well compared to the previous SOTA model.