The impenetrable nature of optical signal propagation renders visible-light communication (VLC) a widely adopted communication method for confidential-information transmission. Moreover, the advantages of green environmental protection, no electromagnetic interference, and the ability to alleviate the shortage of spectrum resources allow the applications of VLC to expand from indoors to hospitals, mines, underwater, etc. However, the limitations of light propagation result in visible light being adversely affected by obstacles. The collision of beams with obstacles creates varying degrees of optical-power loss. Additionally, the visible-light positioning (VLP) performance achieved by relying on the received signal strength (RSS) deteriorates. In recent years, numerous studies pertaining to visible light have been performed under obstacle environments; however, most of them consider the human body as the main research object and simulate it as a three-dimensional or planar shape as a general model of obstacles, which is not adequately precise to express the characteristics of different obstacle types. Therefore, this study extracts the digital features of obstacles based on point cloud technology to realize the generalized expression of obstacles and proposes a quantitative analysis method for point cloud shadows, based on which a visible-light shadow channel model with high applicability is established. In this study, an experimental site was established to validate the point cloud shading analysis method and the reliability of the point cloud channel model. Simultaneously, the proposed model was used to simulate and analyze the shading of optical signals and power loss under different link propagations. A two-layer BP neural network optimized using a genetic algorithm (GA-BP) was applied to realize fingerprint positioning, and the effect of obstacles on the VLP performance as well as possible solutions were analyzed.

First, a depth camera was used to scan obstacles from different angles to obtain a multi-slice point cloud, and a complete obstacle point cloud model was generated using iterative closest point (ICP) algorithm registration. To reduce the time cost and the complexity of the algorithm, the complete obstacle point cloud must be downsampled. To avoid destroying the structural features of the obstacle, we adopted the voxel downsampling method. After performing the two steps above, the point cloud data were preprocessed completely and obstacles were placed in the conventional channel space via a coordinate-system conversion to complete the quantitative analysis of shadows. The core idea of point cloud shadow quantization is to calculate the intersection point of the light link passing through obstacles and falling on the receiving plane in space based on the principle of light propagation along a straight line, which is the shadow point. Subsequently, the convex hull of this shadow point set, which is a rough shadow area, is obtained. The surfaces of some obstacles in the actual environment are not completely closed; therefore, after the convex hull is obtained, one must determine whether the distance between the receiver coordinate point and the nearest neighbor in the set of shadowed points satisfies the spacing requirements. When the receiver coordinate point is located inside the convex packet and fulfills the conditions of spacing judgment, the optical link is obstructed by obstacles and the optical signal cannot reach the receiver. Thus, the shadowing coefficient is set to 0; otherwise, it is set to 1. A point cloud shadow channel model can be established by combining this coefficient with the channel DC-gain formula. In this study, a 2.6 m×2.6 m×2 m experimental site was established to verify the reliability of the shadow channel model developed based on the point cloud shadow-quantization method. The signal-power value of the experimental ground was measured, the simulated power value of the experimental environment was calculated based on the proposed model, and the power-distribution maps of the two environments were obtained for comparison. Finally, an application analysis of the proposed model was performed by predicting the positioning results using the GA-BP network in the simulation space.

Based on the experimental-site measurements and simulation calculations, the shaded regions in the power-distribution maps are highly identical (Fig. 4). Meanwhile, the results of error calculation show that the average error of the normalized power values in the two shaded regions is only 0.0277, whereas the maximum error is 0.1671 (Fig. 5). The experimental results indicate that the point cloud quantization analysis method is effective; therefore, we simulated the power distribution with a higher density of receiving points in a 2.6 m×2.6 m×3 m simulation space based on the point cloud shadow channel model. The results show that owing to obstacles, the average total received power in the shadow area decreases from 6.52×10^-6 to 3.88×10^-6 W, the average contribution ratio of direct power decreases to 23.92%, and the power decreases to 0 W in most areas. Although the primary reflected power is similarly reduced (with a minimum value of 1.51×10^-6 W), the average contribution ratio increases to 76.08%, which mitigates the substantial power loss (Table 3). In the fingerprint positioning application based on the proposed model, the root-mean-square error (RMSE) reaches 20.82 cm when the direct power is used as a feature, and the maximum error is 2.67 m (Table 5). Despite the increase in data, the prediction effect of the target points in the shaded area still exhibits significant errors. By contrast, when combining the reflected power for zonal positioning, the RMSE reduced to 1.58 cm, and the maximum error is only 13.58 cm, which improved the positioning performance considerably.

In this study, we focused on the effect of obstacles, which is a typical feature in complex environments, on visible-light signal propagation using point cloud technology. Based on point cloud simulation of obstacles to obtain their specific three-dimensional coordinates in the channel space as well as the principle of light propagation, a point cloud shadow quantization analysis method was proposed. Additionally, based on the conventional indoor VLC channel model fully integrated with obstacles to achieve light obstruction, we deduced the channel-gain formula. The effectiveness of the point cloud shadow-analysis method was verified on a 2.6 m×2.6 m×2 m experimental platform. Differences in the shadow areas formed by optical signals propagating in different links as well as the degrees of direct and reflected power losses were analyzed via a simulation based on the proposed model. Through the model application of visible-light fingerprint positioning, the adverse effect of obstacles on the positioning performance was analyzed. The RMSE was 20.82 cm when only direct power was used, whereas it was 1.58 cm when the partitioned positioning method combined with reflected power, as proposed herein, was adopted. The point cloud channel model established in this study provides an effective method for communication and positioning research in an environment with obstacles, and its application value is significant.