Parking Environment Obstacle Recognition Based on Ultra-Wideband Radar and Multi-Feature Fusion Transformer

Ultra-Wideband(UWB) digital keys are widely used in smart vehicles, and vehicle-mounted UWB anchors can switch to radar mode to enhance environmental perception and safety. However, applying UWB radar for obstacle recognition in the parking environment faces challenges such as limited velocity sensing range, short signal duration, and dispersed signals. To address these issues, this paper proposes a multi-feature fusion Transformer model based on a self-attention mechanism for micro-Doppler signal recognition and classification. Fast Fourier Transform (FFT) and Short-Time Fourier Transform (STFT) are used to extract velocity and time-velocity features, which are then fused and input into the Transformer model for deep learning training. Experimental results on real parking environment data show that the proposed model outperforms a Long Short-Term Memory (LSTM) model using only STFT velocity features, demonstrating its effectiveness and feasibility in the parking environment.