Research on Train Tracking and Recognition Algorithm for Complex EMU Depots Based on Multi-modal Fusion

Aiming at the problems of poor environmental adaptability and incomplete coverage of vehicle types in the single video recognition scheme commonly used in the current Control Centralized System (CCS) of EMU depots, this paper proposes a train tracking and recognition algorithm based on the fusion of millimeter-wave radar, LiDAR, and multi-view video images. The algorithm constructs a “speed-structure-texture” trinity cross-modal feature system, breaks through the limitation of traditional schemes that only adapt to EMUs, and realizes type classification, precise position tracking, and attribute recognition for both EMUs and ordinary-speed trains. Compared with the single video recognition scheme, the comprehensive recognition accuracy of this scheme in complex scenes is increased by 37.2%, The statistical error of carriages for ordinary-speed trains is no more than 0.3, and the body number recognition accuracy reaches 98.7%, providing core technical support for the upgrading of CCS systems to mixed train scenarios.