Hardware Software Co-design of an Embedded RGB-D SLAM System

Vision sensors with color and depth hâve recently gained popularity. Autonomous ve- hicles benefit from new 3D perception methods thanks to these sensors. We hâve investigated the various processing stages of the System to make contributions at the sensor-algorithm coupling and computing architecture levels. This study began by conducting a thorough experimental analysis of the impact of sensor acquisition modalities on locali­ zation accuracy. We introduced RGB-D HOOFR- SLAM, which was assessed using a self-collected RGB-D dataset. We compared the measurement results to those of the most advanced algorithms. The results revealed a significant réduction in lo­ calization errors and a significant gain in proces­ sing speed compared to the state-of-the-art stéréo and RGB-D algorithms. We proposed the implé­ mentation of the HOOFR SLAM front-end on an FPGA-based architecture. This innovative archi­ tecture delivers superior performance and a trade- off between power consumption and processing time.