To address the challenges faced by epidemic-prevention robots operating in complex indoor multi-room environments with dense human presence, this work proposes a global path planning algorithm termed Bidirectional Target Bias APF-RRT* (BTB-APF-RRT*). The approach employs morphological closing operations from image processing to preprocess the environment map, thereby extracting inter-room connectivity boundaries and optimizing connected regions to reduce ineffective search spaces. By integrating the bidirectional RRT* framework, the algorithm significantly accelerates convergence, while the incorporation of a bidirectional target bias strategy and artificial potential field guides the expansion of random search trees, effectively mitigating randomness and blind exploration. Simulation results conducted in multi-room indoor scenarios with high human density demonstrate that, compared with existing state-of-the-art algorithms, the proposed method achieves notable improvements in terms of sampled nodes, planning time, and path length, thereby exhibiting superior convergence efficiency and path optimization performance.