We present a guiding system to help blind people walk in public spaces while making their walking seamless with nearby pedestrians. Blind users carry a rolling suitcase-shaped system that has two RGBD Cameras, an inertial measurement unit (IMU) sensor, and light detection and ranging (LiDAR) sensor. The system senses the behavior of surrounding pedestrians, predicts risks of collisions, and alerts users to help them avoid collisions. It has two modes: the "on-path'' mode that helps users avoid collisions without changing their path by adapting their walking speed; and the "off-path'' mode that navigates an alternative path to go around pedestrians standing in the way. Auditory and tactile modalities have been commonly used for non-visual navigation systems, so we implemented two interfaces to evaluate the effectiveness of each modality for collision avoidance. A user study with 14 blind participants in public spaces revealed that participants could successfully avoid collisions with both modalities. We detail the characteristics of each modality.