The client is US based manufacturer of low-power semiconductor. They solve customer problems across the network, from Edge to Cloud, in various domains. They enable businesses to create a smart, secure, and connected world using innovative technologies. They now want their hardware platform to be capable of running CNN (Convolutional Neural Network) AI models at faster speeds while using very little power and maintaining a reasonable frame rate. The client approached Softnautics to help them develop an AI model running on an edge device with a camera linked to the FPGA that can categorize and classify different vehicles. The challenges of deploying a convolutional neural network on an edge device is intricated. Tasks like dataset preparation, image pre- and post-processing, neural network architecture selection, CNN model training and fine tuning, needs to be performed as the hardware should be compatible with the neural network architectural layers
- Achieved vehicle classification model accuracy as high as 95%
