Font Size: a A A

Research On Approaches To Adaptive High-Dynamic-Range Vision Towards The Automotive Applications

Posted on:2015-05-12Degree:DoctorType:Dissertation
Country:ChinaCandidate:S D PanFull Text:PDF
GTID:1108330479979665Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
Grounded on the National Natural Science Foundation grand project "Key Tech-niques and Integrated Validation Platform for Autonomous Land Vehicle" and the Na-tional Natural Science Foundation key project "Research on Key Scientific Problems in Intelligent Vehicle Driving on Highway", a high-performance on-board camera has been developed, and on this basis, the methods for acquiring and preprocessing high dynamic range road images are proposed. The reliability and stability of the road image details under different lighting conditions are effectively improved, which potently guarantees the stable and reliable scene understanding in structured environment. The main contri-butions and innovations of this dissertation are as follows:First, a high-performance on-board camera has been developed, which significantly improves the adaptability to the variation of the lighting condition of the scene. Studied the special demands of the automotive applications of the imaging devices in structured environment, an on-board CMOS camera with adjustable dynamic range has been de-signed and implemented, whose maximum dynamic range is up to over 150 dB.Second, a method for acquiring high-dynamic-range road images based on the con-trol of the multi-slope response is proposed, with which the reliability of the image details and the adaptability to the scene illumination range are simultaneously achieved. Firstly, the mathematical model of the multi-slope response curve of the image sensor is estab-lished. Secondly, accurately estimated the scene illumination range, a solution for the parameters of the multi-slope response curve is designed and implemented. The reliabil-ity of the details of the acquired images is ensured using the proposed method while the scene illumination range does not exceed the dynamic range limit of the camera.Third, an edge-preserving image iterative smoothing algorithm is proposed, with which not only the edge distortions are effectively avoided, but also the computational efficiency is greatly improved. Firstly, according to the concept of the edge-preserving image smoothing techniques, an efficient asymmetric anisotropic diffusion algorithm is implemented. Secondly, using the spatial differentials of the observed image as the dis-tance measures of the edge-stopping function, blocking artifact caused by the inverse diffusion is effectively avoided. Experiments prove that the efficiency of the proposed algorithm is two orders of magnitude faster than the standard bilateral filter.Fourth, a constant-time bilateral filtering algorithm based on local histogram is pro-posed, with which the efficiency of the road image preprocessing is further improved. Firstly, a multi-box bilateral filtering algorithm is designed, in which the spatial filtering kernel is sparsely approximated using a fixed number of box filters. Secondly, by virtue of the constant-time integral histogram technique, the computational complexity of the pro-posed algorithm is reduced to O(N). When approximating the standard Gaussian bilateral filter, high accuracy can be achieved using the proposed algorithm with few box filter-s, and its performance is more robust with respect to the values of the input parameters, compared with other congeneric algorithms.Fifth, a preprocessing method for the high-dynamic-range road images based on "il-lumination+reflectance" decomposition is proposed, with which the reliability and stabil-ity of the image details are significantly improved. Firstly, an edge-preserving algorithm for estimating the scene illumination is designed. Secondly, a high-dynamic-range road image is decomposed into a high-dynamic-range illumination map and a low-dynamic-range reflectance map. Finally, the dynamic range of the illumination is compressed ac-cording to the illumination range of the scene. Experiments prove that the effects caused by the non-uniformity of the scene illumination distribution and the changes of the illu-mination range are effectively reduced using the proposed method.The above achievements are the part of the key techniques for the autonomous Hongqi HQ3 car to perceive and understand the structured environment, which not only meet the needs under conventional lighting conditions, but also provide stable and reliable road im-ages for the environment perception and understanding module in real time in the scenes under complex lighting conditions, such as when the vehicle is passing a tunnel. They have provided a reliable guarantee of the road information for the first successful long-distance (286 km) automatic driving test on highway under normal traffic conditions in China.
Keywords/Search Tags:high-performance on-board camera, high dynamic range im- age acquisition, edge-preserving image smoothing, constant-time bilateral filter, illumination-reflectance decomposition, road image preprocessing
PDF Full Text Request
Related items