Abstract: Owing to the localization error associated with traditional SLAM frameworks used in dynamic scenes, the established scene dense map contains dynamic objects with overlapping motions, leading to insufficient robustness of localization and mapping. For indoor dynamic scenes with humans as the main dynamic object, and from the perspective of "temperature", we propose a collaborative scheme of multi-sensor SLAM using a combination of the thermal imaging system and depth camera to tackle the issue of positioning and mapping in indoor dynamic scenes. First, we establish a set of joint calibration strategies for the thermal imager and depth camera, redesign the calibration plate and scheme to complete the internal parameter calibration, external parameter calibration of the camera and image registration, and obtain the one-to-one corresponding three-mode images of RGB, depth, and heat (RDH). Second, we use the thermal image to obtain the mask image of the human body. Then, we construct the static feature extraction and data association strategy under