A Near-Sensor Processing Approach with SDFT Kernel for CMOS Image Sensor Based Camera System

A near-sensor analog processing based approach with sliding discrete Fourier transform (SDFT) kernel has been proposed in this paper. A complex valued kernel based on the SDFT has been introduced for edge detection operation. This kernel is implemented in the form of resistors, which convolve with incoming raw data from a CMOS image sensor. The SDFT based kernel performs better in terms of egde connectivity, removal of irrelevant edges, and noise. This framework has been implemented and tested using LTspice for circuit simulations and Matlab for performance evaluation of the outputs. The performance evaluation of the proposed SDFT approach in-terms of accuracy is better than the Prewitt and Sobel. Near-sensor analog processing approach has reduced the datasize for transmission and storage by 90%. Furthermore, this approach does not require the usage of an analog to digital converter (ADC) during the processing. The near-sensor analog processing circuit consumes 200 mW power on average for the tested image data.