Kamera görüntüleri ile yol eğimlerinin bulunması

Lateral and longitudinal control of a vehicle are the basis of autonomous driving. Road slopes (parameters) are vital information that are used to control a vehicle laterally in lane keeping assistance and longitudinally in adaptive cruise control respectively. We present four methods estimating road slopes or 3D road parameters from instant road images using a front monocular camera. We start with a simple algorithm Geometry Based (GB) method that extracts invariant properties on 2D image. The second solution Covariance Based (CB) method, which uses covariance matrix properties of 2D road lines as features, is similar to supervised learning techniques. We present a systematic road slope estimation called as Iterative Least Squares with Optimization (ILSO) in our third solution. The algorithm is smartly divided into two processes as Least Squares (LS) and optimization depending on their estimation capabilities. LS estimates road parameters and optimization step estimates longitudinal distances. It is an iterative estimation that for each iteration LS and optimization run one time. It is an easy technique due to its systematic character that it does not require complex geometric calculations and match between left and right line. In our final method Local Features Based (LFB), we use the SIFT local features between two consecutive image pairs to estimate road slopes. GB, CB and ILSO methods rely on road lines and provide in advance estimations that enables proactive driving, whereas LFB uses the entire image and estimates road slopes when the road slope actually changes at that moment. All our methods are analytical and give explicit results except CB, which predicts a parameter by choosing a minimum metric. We use two different road models to characterize 3D road parameters for the algorithms. Multi-Layer Perceptron (MLP) learning method is used as a benchmark in our tests. We compare the performances and system requirements of our methods under controlled and uncontrolled test environments.