Real-time PTV system implementation on multi-SoC architecture accelerated by OpenCL

Measuring discrete particle trajectories in the air and monitoring airflow movement through 3D particle tracking technology (3D PTV) have numerous applications in smart homes, environments, energy, and other fields. In this study, an intelligent instrument for a real-time 3D PTV system is designed and developed based on the data flow streaming model. A high-level set of various functions is implemented following the client-server architectural model to provide services like 3D tracking and camera calibration. The model is deployed on several master-slave-based SoC FPGA acceleration boards to meet strict constraints like a high frame rate required for high trajectory precision. A functional decomposition of the 3D tracking service is elaborated to map particle detection and temporal tracking processes on three slave boards, one per camera. The remaining processing (spatial matching and 3D reconstruction) and the client requests management are mapped on the master board. On the FPGA processors of slave boards, treatment has been accelerated with a pipeline structure of the internal processes interleaved by FIFOs (First-In, First-Out) with the help of OpenCL. Experiments have been conducted using low-cost Intel DE10 standard boards.