Road understanding using single level wavelet package approximately compressed sensing

In process of outdoor autonomous mobile robot visual based road understanding, the vision algorithms are not so robust, and a big data is encountered in the real time process of navigation images, which are all led to by shadows, causing cracks and other singular signals encountered. To resolve these difficulties, the Single Level Wavelet Package Compressed Sensing (SLWPCS) algorithm is proposed. Combined with the Adaptive Genetic Algorithms (AGA) based image segmentation algorithm, a real time road understanding algorithm system is established. Through coarse measured approximation road image by wavelet packet decomposition of multi-levels, the best scale space not affecting the "road-non-road" second classification is determined. Road image is decomposed by using sym8 wavelet in the best scale space. A compressed sensing matrix is used to realize dimension reduction of the high-frequency coefficients in addition to the diagonal direction coefficients. OMP algorithm is used to reconstruct high-frequency coefficients. And together with the low-frequency coefficients, the reconstruction of the original image is finished. The fitness function is constructed by the gray value largest variance between classes and the minimum variance within a class for each image frame. The optimal adaptive threshold segmentation is then realized, and the road boundaries can be found.A wheeled autonomous land vehicle is selected as the research platform, and the algorithm is tested by the actual roads and the outdoor path driving video of the mobile robot provided by CMU. The experimental results show that this method can detect the boundaries robustly under varying road conditions including shadow, crack, and illumination change. The real-time performance can be well satisfied.