Analytical Assessment of Heterogeneous Computing Architectures for Real-Time Video Analytics at Nigerian Border Checkpoints

Automated video analytics for vehicle and pedestrian screening at Nigerian border checkpoints and immigration control points could significantly improve throughput and detection effectiveness, yet the computational architectures best suited for real-time performance under the power and environmental constraints at remote border posts have not been analytically assessed. This study analytically evaluated CPU, GPU, FPGA, and NPU heterogeneous computing architectures for real-time video analytics workloads representative of border checkpoint applications: vehicle licence plate recognition, facial comparison against watch-list databases, and thermal camera fever screening. Benchmark workloads were implemented in TensorFlow and OpenCV and evaluated on NVIDIA Jetson AGX Xavier (NPU+GPU), Intel Neural Compute Stick 2 (NPU), Xilinx Alveo U50 (FPGA), and a desktop Intel Core i9 (CPU reference). Performance metrics included inference throughput (frames per second), accuracy, power consumption (watts), and unit cost. NVIDIA Jetson AGX Xavier achieved the best balance of performance (47 fps for combined three-task pipeline) and power efficiency (18W average), with the highest cost at 380,000 naira. Intel NCS2 achieved 12 fps at 1.5W, suitable for single-task deployments with severe power constraints. FPGA delivered 38 fps with 11W but required significant development investment. The study recommends Jetson AGX Xavier for high-throughput manned border crossings and Intel NCS2 for solar-powered remote crossing points. Collaboration with the Nigeria Immigration Service for a pilot deployment is proposed.

Keywords: heterogeneous computing, video analytics, border checkpoint, NVIDIA Jetson, Nigeria Immigration