Partial Scene Reconstruction for Close Range Photogrammetry Using Deep Learning Pipeline for Region Masking

3D reconstruction is a beneficial technique to generate geometry of scenes or objects for various applications such as computer graphics, industrial construction, and civil engineering. There are several techniques obtain the an object. Close-range photogrammetry inexpensive, accessible approach obtaining high-quality object reconstruction. However, state-of-the-art software systems need stationary scene controlled environment (often turntable setup with black background), which can be limiting factor scanning. This work presents method that reduces allows capture multiple independent motion. We achieve this by creating preprocessing pipeline uses deep learning transform complex from uncontrolled into background then fed existing Our achieves using models detect track through scene. The detection tracking semantic-based supports available pretrained custom networks. develop correction mechanism overcome some shortcomings, namely, object-reidentification detections same show effective address motion reconstructed limited no knowledge camera environment.