This paper presents a method for humanoid robots to perform complex loco-manipulation tasks in industrial settings by combining fast dense 3D tracking with visual SLAM (vSLAM) using wide-angle depth images. The approach integrates a task-space whole-body optimization control framework that enables robots to manipulate and assemble large-scale objects while walking. The method uses a new fast dense 3D model-based tracking algorithm with a wide-angle depth camera, combined with vSLAM for simultaneous localization and mapping. This allows the robot to track objects while moving, even when the object is moving relative to the robot. The system was tested with two different humanoid robots, HRP-2KAI and HRP-5P, in scenarios involving the manipulation of large objects such as wheelbarrows and bobbins. The experiments demonstrated the effectiveness of the approach in real-world industrial settings, where the robot successfully performed tasks such as rolling and assembling a heavy bobbin. The method uses a combination of visual tracking and vSLAM to accurately localize the robot and the manipulated object, even in challenging environments with limited texture or moving objects. The system is open-source and can be adapted to various humanoid robots with minimal programming effort. The results show that the approach is robust and efficient, enabling humanoid robots to perform complex loco-manipulation tasks in industrial manufacturing.This paper presents a method for humanoid robots to perform complex loco-manipulation tasks in industrial settings by combining fast dense 3D tracking with visual SLAM (vSLAM) using wide-angle depth images. The approach integrates a task-space whole-body optimization control framework that enables robots to manipulate and assemble large-scale objects while walking. The method uses a new fast dense 3D model-based tracking algorithm with a wide-angle depth camera, combined with vSLAM for simultaneous localization and mapping. This allows the robot to track objects while moving, even when the object is moving relative to the robot. The system was tested with two different humanoid robots, HRP-2KAI and HRP-5P, in scenarios involving the manipulation of large objects such as wheelbarrows and bobbins. The experiments demonstrated the effectiveness of the approach in real-world industrial settings, where the robot successfully performed tasks such as rolling and assembling a heavy bobbin. The method uses a combination of visual tracking and vSLAM to accurately localize the robot and the manipulated object, even in challenging environments with limited texture or moving objects. The system is open-source and can be adapted to various humanoid robots with minimal programming effort. The results show that the approach is robust and efficient, enabling humanoid robots to perform complex loco-manipulation tasks in industrial manufacturing.