A new method for vehicle detection using MexicanHat wavelet and moment invariants

Considering the limits of power and processors in wireless intelligent terminals for vision-based intelligent transportation systems, a new strategy is proposed for vehicle detection in this paper. The contribution of the proposed strategy consists of the adaptive edge detection and feature extraction. In the step of the adaptive edge detection, an adaptive threholding rule is proposed and used for the wavelet modulus maximum method. The threshold rule keeps the edge sharpness and the robustness to noise to improve the detection accuracy. To reduce the computation time, the moment invariants based on Hu moments are calculated to reduce feature data and keep the feature with the invariance of translation, scale and rotation. The real vehicle and non-vehicle images are used to construct the database for experiments. The experimental results show that the average correct rate is about 85%-88% while the running time is less than 1 second. The proposed strategy keeps the balance of the accuracy and running time to a certain extent.