Recently, the use of assistive vehicles in industrial or daily day tasks started to grow rapidly. Therefore, it is important to guarantee safety to the robot and to any other moving element in the environment (either people, animals or other robots). In this work, we develop and implement a navigation assistive system based on collision risk estimation using depth sensors. Speed and steering constraints are applied to semi-autonomous assistance vehicles to avoid hazardous situations and to improve the users welfare. We calculate a collision risk indicator based on the tracking of moving elements from the scene, by means of a visual tracking approach and a proposed motion model. The performance of the system is tested in selected situations. Furthermore, the motion model associated with people is empirically validated. Finally, the simulation results included here, show the effectiveness of the system in reducing the imminent collision risk up to 90%, without imposing drastic decisions over the vehicle movement.