순차적 최적화 방법을 사용한 스틱셀 파라미터 튜닝
이선영, 서재규, 정호기 · 한국자동차공학회 추계학술대회 및 전시회 · 2014
Stixels are rectangular sticks utilized for representing obstacles in driving environments. Stixel generation procedure consists of two main stages: free space computation and obstacle height estimation. In these stages, several essential parameters are used to control the behavior of the algorithms. Thus, those parameters should be tuned to achieve better performances. Although evaluating all parameter combinations is the best way to find their optimal values, it is unrealistic in terms of computational costs. To overcome this problem, this paper proposes a Stixel parameter tuning method based on a sequential optimization. Assuming that the parameter space has a single mode, the proposed method repetitively conducts the procedure that finds the optimal value of one parameter while fixing the other parameters until the performance converges. Experimental results show that the proposed method can properly tune the parameters used in the Stixel generation procedure with a feasible computational complexity compared with the method that evaluates all parameter combinations.
Cited by 0 works.