Visible light communication and positioning (VLCP) systems have been rarely researched to sense and model the obstacles. However, the accumulating database containing the signal amplitudes and receiver positions can be exploited to detect and reconstruct the positions and shapes of the obstacles. In this work, we propose an algorithm to model the obstacles in a VLCP system, where several light-emitting diodes (LEDs) are set at known positions, and the users equipped with photodiodes (PDs) randomly move to receive the LEDs' signals at different positions. Based on the received signals, the blocked links between the users and LEDs can be distinguished, and thus we can reconstruct the obstacle information. Simulation and analytical results show that the algorithm can estimate the positions and shapes of obstacles. Such an algorithm expand the services of the VLCP systems, and can sense the three-dimensional (3D) information of the environment through the light channel coefficients so as to circumvent the expensive devices, such as laser radars. The proposed algorithm can also support navigation and prediction of wireless channel fading.