Graphics pipeline optimization
Platform: NVIDIA Jetson Xavier
Description: For a customer product with multiple camera inputs, graphics processing and output rendering, Embien worked to optimize the entire pipeline to achieve very short processing time. Re-writing some of the core drivers, removing unnecessary memory copy operations and enabling direct memory access by GPUs, we were able to achieve very small glass to glass latency.
Algorithm Porting to CUDA
Platform: NVIDIA Tegra TX2
Description: With rich experience in algorithm development, Embien reorganized some algorithms written in C and Python to CUDA, leveraging the massive number of GPUs available in the Jetson platforms. Our team carefully studied the algorithm provided by the customer and restructured to CUDA kernels in order to utilize the parallel execution capability and significantly improved the system performance.
360° Object Tracking Image Stitching
Platform: NVIDIA Jetson Nano
Description: A triple-camera based device design with 360 ° field of view for commercial buildings and hospital surveillance. It includes intelligent object tracking with image stitching solution. Wide 360 ° FOV is provided by stitching the frames from at least 3 cameras’ placed at 60 degree each. OpenCV SIFT algorithm is used for stitching images and deep learning algorithms were used to detect the face or object in the frame with support for USB based joystick.