The implementation of computer vision algorithms in fixed point

Abstract

For many original foreground detection algorithms, they are all first simulated using floating point algorithm and later transformed into fixed point to implementation on some hardware, as DSP or FPGA. However, these designs and processes take much time than those float point designs. Furthermore, the word-length in fixed point implementation is the key point. Different choice of the word-length has to lead different trade-off between foreground detection quality and implementation complexity.