Comparative Evaluation of 3D Fire-Source Localization Technologies for Ship Engine Room Applications

3D fire-source localization is important for automated fire monitoring in ship engine rooms, but comparisons of range-information acquisition approaches remain limited, particularly under real-fire conditions. This study compared RGB-D-based methods (C1 and C2) and camera–LiDAR association-based methods (L1, L2, and L3) using the same fire-detection results and a common coordinate frame. The surrogate-target, real-fire, and range-limit experiments were conducted in a ship engine room, a separate controlled test environment, and a corridor environment, respectively. Performance was evaluated by output availability (based on fire-detected, synchronized frames), geometric accuracy, temporal stability, and physical association. In the surrogate-target experiment, RGB-D methods produced no valid coordinates beyond 6 m, whereas camera–LiDAR methods produced coordinates throughout the range. Under real-fire conditions, camera–LiDAR methods maintained at least 97% output availability, but median 3D localization errors were 111.9–181.1 cm with positive longitudinal biases of 85.7–158.6 cm. Background-reference analysis showed differences of 3.5 cm (L2) and 5.2 cm (L3) between longitudinal biases and background-structure positions. L3’s temporal smoothing reduced frame-to-frame variation without eliminating this systematic spatial bias. Fire-detection rates and L3 output availability remained high over 11–12.6 m. These results show that high output availability does not guarantee accurate localization or correct physical association and that these four performance aspects should be evaluated separately.