Real-time evaluation of different indexes in precision agriculture using a heterogeneous embedded system

•A Real-time processing algorithm and their embedded implementation have been proposed to monitor vital parameters such as NDVI and NDWI using multispectral images from agricultural fields based on robots and UAVs.•A H/S Co-design approach has been treated in order to propose an optimal embedded implementation based on OpenMP and C/C++ in several embedded systems.•The implementation has been developed and the constraints have been discussed. In this work, we present a real-time embedded implementation of an algorithm dedicated to monitoring agricultural fields. This algorithm is based on normalized indices, such as the Normalized Difference Vegetation Index (NDVI) and the Normalized Difference Water Index (NDWI). The problem of most algorithms in this context is real-time processing, especially when we talk about applications that require time precision. The proposed implementation is based on the application of a Hardware/Software Co-design approach. For the embedded platform, we used the heterogeneous system contains CPU and GPU type XU4 and TX1. In this context, we used the parallel programming language OpenMP to have an optimal embedded implementation. The results showed that we could process 66 images/s using a desktop, 20 images/s in the XU4, and 17 images/s for TX1.