CEVA-SLAM SDK is targetted at the CEVA-XM family of intelligent vision DSPs and the NeuPro family of AI processors. The SDK incorporates the hardware, software and interfaces necessary to lower the entry barrier for companies looking to design efficient SLAM implementations into low-power embedded systems.
The CEVA-SLAM SDK includes a detailed interface from a CPU to offload the heavy lifting SLAM blocks to the CEVA-XM DSP. The SLAM blocks use the DSP efficiency to support fixed point and floating point math and extend the device’s battery life. The SDK includes capabilities for image processing (including Feature Detection, Feature Descriptors, Feature Matching), Linear Algebra (including Matrix Manipulation, Linear Equation Solving), Fast Sparse Equation solving for Bundle Adjustment and more.
The CEVA-SLAM SDK running a full SLAM tracking module on the CEVA-XM6 DSP at 60 fps consumes only 86mW. When deployed with a CEVA-XM DSPs or a NeuPro AI processor, customers can address a broad range of use cases and applications requiring SLAM, such as visual positioning, along with classical and Neural Network workloads for imaging and vision.