Efficient architecture for circle detection using Hough transform

Abstract

This paper presents an efficient architecture for circle detection using Hough transform. The architecture adopts the scanline-based ball detection algorithm for the edge detection stage and edge-flag algorithm for the voting process. To bolster the performance of the voting process, when drawing a circle, we divide it into 16 sub-parts and compute the parts in parallel. The proposed design employs an internal memory block for the edge list. Our simulation results show that the benchmark images in the VGA size of 640 x 480 are processed within 10ms, which indicates that the proposed architecture can satisfy the speed requirements of most real world applications.

FPGA-Based Real Time Embedded Hough Transform Architecture for Circles Detection

“…Here, we present the highlights found, a survey of Hough transform methods can be read in [2] and [13]. We can notice some facts: 1) all the FPGA implementations [1], [3], [19], [16], [8], [11] use embedded internal memories or BRAMs (Block RAMs), 2) all the works consider only the acceleration of the voting process of the Hough transform. In this process the goal is to accumulate intersection points in the Hough parameter space, corresponding to the number of possible circles.…”

“…Then a vote is done to find the local maximum that are considered as real circle. This process is memory and time consuming, two different approaches emerged, the first approach uses the original Hough transform algorithm were the parametric equations of a circle are used to find the center and the radius of the circles [1], [3], [16], [8], [11] and the second approach uses a modified version of Hough transform called One dimensional Hough transform algorithm [19]. We describe below the works related to both approaches.…”

“…In [16], authors adopt the scanline-based ball detection algorithm for the edge detection stage and edge-flag algorithm for the voting process.…”

“…Zhou and al. [19] Steps [15][16][17][18][19][20][21][22][23][24] of algorithm 1, the more voted distance r become the radius of the center candidate (i, j); 3) If the value V r (i, j, r) is greater than f * 4 √ 2r then ||(x, y) − (i, j)|| = r becomes a circle. Steps [20-22] of algorithm 1. f can be adapted as the sensitivity threshold to detect circles.…”

“…We have validated our algorithm on image of closed contour as show in figure 1. In this image f correspond to the sensitivity threshold visible in step [16] of our algorithm 2. A green circle indicates that we have localized a closed contour.…”