Adaptive Traffic Lane Detection Based on Normalized Power Accumulation

Precise traffic lane boundary identification is the precondition for a good performance of the side-looking traffic flow detection radar. Some approaches have been proposed and developed for this specific purpose. However, these methods either require onerous measurements and prior knowledge, or large amounts of statistics and a long learning duration which cause great difficulties on the practical application of the device. Motivated by this, we propose a novel practical lane boundary estimation algorithm. The proposed method is on the basis of normalized power accumulation. In this method, an adaptive threshold is set to detect the vehicles in the detection region of the traffic radar as well as a vehicle filter to select the vehicle having the strongest echo power. Then the power distribution of the selected vehicle is normalized and put into the power accumulation procedure. Finally, traffic lane boundaries are identified on the basis of the distribution of the normalized power accumulation. Some of the previous approaches are also analyzed in this paper. To test the performance of the proposed algorithm, a great deal of field tests in different locations is conducted and the test results as well as the comparisons with the previous work are presented.