Visual SLAM systems often ignore the camera’s intrinsic kinematic properties during pose calculation and optimization, leading to degraded localization accuracy and sparse data associations. To address these limitations, we propose KSLAM: a tightly coupled visual SLAM framework that integrates rigid-body kinematics into prior pose calculation, data association, and nonlinear optimization. KSLAM calculates the camera's Rigid-Body Kinematics parameters directly from image sequences in real time. It predicts a kinematically consistent prior pose for the subsequent frame. Compared with traditional constant-velocity models, our method demonstrates superior robustness under high-speed motion, using limited computational resources. To improve data association, we introduce a motion-adaptive strategy which achieves higher number of reliable matches. It leverages calculated translational and rotational motion to dynamically select pyramid levels and construct elliptical search regions. Furthermore, we consider kinematic consistency constraints in pose error modelling, formulating a tightly coupled optimization problem solvable by the LM algorithm. This formulation mitigates issues such as local minima caused by inaccurate priors and sparse correspondences, thereby improving tracking stability and trajectory precision. Extensive experiments across multiple public datasets demonstrate that KSLAM significantly improves pose accuracy and data association performance while maintaining a lightweight design suitable for real-time applications.