Name
Abstract | ||
|---|---|---|
Due to high complexity on matching computation, real-time object tracking is generally a very challenging task for practical applications. This paper proposes a new algorithm for moving object tracking, which improves the traditional KLT algorithm by using the motion information for feature points selection to avoid the irrelevant feature points residing in the background area. Moreover, this paper designs the hardware architecture of the FPGA part to accelerate the computation by optimizing the inherent parallelism of the algorithm. The proposed algorithm is able to significantly reduce the computation time. Experimental results show that our algorithm implemented in an FPGA-SoC (Zynq 7020, 667 MHz) requires only 0.030 s to handle a VGA resolution frame, which is suitable for real-time tracking. This achieves up to (30{times }) performance improvement compared with the desktop PC (i3, 3.4 GHz), or (370{times }) compared with the ARM (Cortex-A8, 1 GHz). The experiment also shows that our approach consumes less energy significantly than PC and ARM for the same workload, which indicates that it is suitable for energy-critical system. |
Year | Venue | Field |
|---|---|---|
2017 | ICA3PP | Deep-sky object,Computer science,Parallel computing,Field-programmable gate array,Tracking system,Video tracking,Computer hardware,Video Graphics Array,Performance improvement,Computation,Hardware architecture |
DocType | Citations | PageRank |
Conference | 0 | 0.34 |
References | Authors | |
7 | 5 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| wenjie chen | 1 | 5 | 2.83 |
| yangyang ma | 2 | 2 | 1.58 |
| Zhilei Chai | 3 | 63 | 13.51 |
| Mingsong Chen | 4 | 279 | 25.10 |
| Daojing He | 5 | 1013 | 58.40 |