Research Projects

The demand for computer vision applications such as surveillance, autonomous vehicles, gesture recognition, and medical applications has led to a growing demand for cameras or CMOS Image Sensors (CIS) with inherent intelligence, low latency, and low power solutions. Conventional CIS are designed to obtain raw images but have been made to operate at low power, under low noise conditions, high speed, and high dynamic range. The specifications of these CIS are determined based on application requirements and various circuit, architecture, and algorithmic level techniques are implemented to achieve the required set of specifications. The major drawback of conventional CIS is that cameras are not smart or intelligent, continuously being operated for an entire duration, increasing total power consumption. Emerging CIS are designed to be intelligent, capable of performing tasks like object classification, detection, feature extraction, and more, increasing their applications in multiple places. The proposed project introduces novelty in design by reducing the power of conventional CIS by incorporating the concept of edge detector, which reduces the number of pixels to be read by the ADC. This reduces the power of the CIS and improves image classification or recognition by sending only few MSB bits instead of the entire default bit resolution of the ADC to convolutional neural networks (CNN). The design incorporates SAR ADC architecture to achieve low power from the ADC. The work will further add hardware-friendly customized CNNs, requiring less resources and enhancing the overall performance of computer vision applications.