Design and Implementation of Blind Assistance System Using Real Time Stereo Vision Algorithms

Abstract

The main objective of this paper aims at designing an efficient blind assistance system for the visually impaired people using real time disparity estimation algorithm. The local window based matching algorithms known as sum of absolute differences (SAD) and zero-mean SAD (ZSAD) are used for the disparity estimation and efficient hardware architectures for those algorithms are implemented in FPGA. The SAD and ZSAD algorithms are implemented for the image resolution of 640x360 pixels with square window of size 8x8 pixels and with a disparity range of 0 to 99. An efficient line buffering scheme for the left and right window of the two camera images is implemented to support the parallel processing mechanism. To provide high frame rate per second (fps) to the artificial vision system, parallel processing architecture for disparity estimation is required, but it consumes huge amount of hardware resources which is not desirable for this application. So, semi-parallel architectures are implemented for SAD and ZSAD algorithms for a compromise between hardware resource utilization and speed. The FPGA resource utilization for the ZSAD and SAD algorithms for a frame rate of 30 fps are 43529 LUTs, 50144 FFs and 34548 LUTs, 37544 FFs respectively. ZSAD algorithm consumes around 30% more hardware resources compared to SAD, but to counter the randomized distortion caused by the non-ideal stereo cameras ZSAD is more preferred. Finally from the disparity, the distance of the nearest obstacle is estimated and blind person is alerted through audio device instructions.