This content is not included in
your SAE MOBILUS subscription, or you are not logged in.
Performance Characterization Platform for Vision Based Driver Safety Systems
Technical Paper
2013-26-0026
ISSN: 0148-7191, e-ISSN: 2688-3627
Annotation ability available
Sector:
Language:
English
Abstract
Many of the Advanced Driver Safety Systems (ADAS) provide vision based driver assistance. All vision based systems are not the same since they differ in algorithms. Thus, it is necessary to compare how robust these different algorithms are and provide the best possible solution to the end user. Such characterization is often difficult due to changing environmental conditions under which the algorithms have to perform. In these systems, performance becomes a critical parameter since any slight lapse could translate to serious danger on road. The reliability of any algorithm depends on its accuracy and consistency under varying environmental conditions. It is important to define the boundary conditions of the algorithm to evaluate these performance parameters. Often one finds out system performance under varying conditions only by expensive road testing.
In this paper, we propose a performance characterization platform to quantify the performance of vision based algorithms for driver safety systems by simulating real-life conditions. With this platform, one can simulate conditions like low light, fog, rain, dust etc. by varying the contrast, brightness, and blur of input image and synthetically adding noise to it. In our platform, one can simulate several different types of noise distributions such as Gaussian, salt and pepper, uniform, and speckle. The accuracy of the algorithm under test is determined for each of these conditions and a plot of accuracy against these parameters is obtained. The plot is useful to define the best operating region in terms of input parameters, for the algorithm under consideration. Additionally, for multiple algorithms, the plots obtained for each algorithm characterizes the performance under identical conditions. Comparison of this performance enables one to determine the best possible solution to give desired accuracy under varying conditions. The newly developed platform saves time as well as money required for extensive road testing.
Authors
Citation
Kharade, P. and Vaidya, V., "Performance Characterization Platform for Vision Based Driver Safety Systems," SAE Technical Paper 2013-26-0026, 2013, https://doi.org/10.4271/2013-26-0026.Also In
References
- Haralick Robert M Performance Characterization in Computer Vision CAIP 1 9 1993
- Forstner W 10 Pros and Cons Against Performance Characterization of Vision Algorithms Proc. of ECCV Workshop on Performance Characteristics of Vision Algorithms 13 29 1996
- Maimone M and Shafer S A Taxonomy for Stereo Computer Vision Experiments Proc. of ECCV Workshop on Performance Characteristics of Vision Algorithms 59 79 April 1996
- Courtney P and Thacker N Performance Characterization in Computer Vision: The Role of Statistics in Testing and Design Imaging and Vision Systems: Theory, Assessment and Applications NOVA Science Books 2001
- Thacker Neil A et al Performance Characterization in Computer Vision: A Guide to Best Practices Computer Vision and Image Understanding 109 3 305 334 2008
- Clark Adrian F and Clark Christine Performance Characterization in Computer Vision a Tutorial http://peipa.essex.ac.uk/benchmark/ University of Essex Colchester, CO4 3SQ, UK 1999
- Ramesh Vishwanathan , Haralick Robert M Professor Department of Electrical Engg. University of Washington, Seattle Performance Characterization of Image Understanding Algorithms PhD dissertation
- Gonzalez Rafael C and Richard E Woods, Digital Image Processing Prentice-Hall, Inc. second 2001
- Kancharla T et al Edge Based Segmentation for Pedestrian Detection using NIR Camera International Conference on Image Information Processing (ICIIP) 2011
- Ahire Shripad , Yarram Naresh Reddy , Mankar Prafull An Image Enhancement and Pedestrian Detection System Indian patent application number:784/MUM/2011