A Lightweight Real-Time Cars and People Detector for Consumer Uncrewed Aerial Vehicles on Edge Platforms

As unmanned aerial vehicles (UAVs) become increasingly prevalent in the consumer electronics market, the demand for low-power, real-time visual perception capabilities, particularly in applications such as smart security, traffic monitoring, and home services, has grown significantly. However, achieving high-precision detection of vehicles and pedestrians on resource-constrained consumer-grade UAV platforms remains a challenge due to high computational overhead and limited energy efficiency. To address this challenge, this study proposes Real-time Detection of Cars and People for UAV (RDCP-UAV), a lightweight real-time object detection model specifically designed for consumer edge devices. The proposed architecture integrates a efficient Multi-Branch Grouping and Reparameterization Aggregation Network (MBGRAN) module, which reduces model parameters and computational complexity while improving multi-scale feature extraction. Additionally, it introduces a Parallel Adaptive Channel and Spatial Self-Attention (PACSSA) mechanism to enhance target feature representation efficiently. Experimental results demonstrate that the RDCP-UAV model achieves excellent detection performance with only 5.30 M parameters and 14.0 billion floating-point operations, significantly lower than those of comparable state-of-the-art methods. Evaluation on the UAVDT dataset verified that the RDCP-UAV model has excellent generalization ability. Importantly, the model delivers real-time inference at 31.74 FPS on the NVIDIA Jetson Xavier NX, a representative consumer-grade edge computing platform, demonstrating its feasibility for deployment and superior energy efficiency in practical UAV systems. This work employs algorithm-hardware co-design to presents an efficient, practical visual perception solution for next-generation intelligent consumer UAVs.