http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Rate Monotonic 스케줄링 알고리즘의 성능 향상을 위한 다중 계층 경쟁 회피 스케줄링 기법
백형부(Hyeongboo Baek) 한국정보기술학회 2018 한국정보기술학회논문지 Vol.16 No.2
A real-time scheduling algorithm for a real-time system is a very important factor because it greatly affects the performance of the entire system. Although the multi-level contention-free policy is the novel technique to significantly improve the performance of existing scheduling algorithms, it has been only applied to EDF (Earliest Deadline First) scheduling algorithm. In this study, we apply the multi-level contention-free policy to RM (Rate Monotonic), and propose a real-time analysis technique for it. Via simulations, we identified up to 53.2% the performance improvement compared to EDF when various numbers of processors and layer heights of multi-level contention-free policy are considered.
경쟁회피 고정 우선순위 스케줄링 알고리즘을 위한 반응 시간 분석기법 연구
백형부(Hyeongboo Baek),백재민(Jeamin Baek) 한국정보기술학회 2019 한국정보기술학회논문지 Vol.17 No.9
The CF(contention-free) policy is a technique to effectively utilize the computing power of multi-processor, which can be incorporated into the most existing real-time scheduling algorithms. Utilizing such the advantage, the CF policy was incorporated into EDF(Earliest-Deadline First) and RM(Rate Monotonic) as well as FP(Fixed-Priority) scheduling algorithms, and DA(Deadline Analysis) as schedulability analysis to support them was also developed. Although RTA(Response-Time Analysis) is better-performing schedulability analysis than DA, it was developed for the CF policy incorporated EDF and RM, not for CF policy incorporated FP. In this paper, we propose RTA for the CF policy incorporated FP scheduling algorithm, and evaluate its performance via self-developed simulator.
SORT와 DeepSORT의 혼합을 이용한 실시간 다중객체 추적
양수진(Sujin Yang),정인화(Inhwa Jung),강동화(Donghwa Kang),백형부(Hyeongboo Baek) 한국정보기술학회 2021 한국정보기술학회논문지 Vol.19 No.10
Deep Simple Online and Real-time Tracking(DeepSORT) is a multi-object tracking technology that improves the accuracy of SORT, and has the characteristic of using a feature map of Convolution Neural Network(CNN) when determining the relationship between objects between consecutive frames in SORT. This feature has the advantage of dramatically improving the accuracy, but has a disadvantage in that the frame per second(FPS) is lowered due to the large amount of additional computing operations required. Therefore, for real-time systems such as autonomous vehicles, SORT has been considered more suitable than deep SORT. In this paper, we propose a method of mixing SORT technology with DeepSORT to remove the shortcomings of the existing DeepSORT. As a result of the experiment targeting MOT-16, a widely used data set used for multi-object tracking performance, it was confirmed that the proposed method dramatically increased the FPS with little loss in accuracy.