Vehicle detection algorithm based on Haar-NMF features and improved SOMPNN

The traditional vehicle detection algorithm based on Haar features and self-organized mapping probability neural networks (SOMPNN) has two shortages: High-dimensional Haar features usually cause long decision time; the constant smooth factor σ of SOMPNN often causes false classification. To solve these problems, low-dimensional Haar-NMF(non-negative matrix factorization) features instead of Haar features and an improved SOMPNN(ISOMPNN) with adaptive smooth factor correction are adopted to build the vehicle detector. First, NMF is used to generate low-dimensional Haar-NMF features. Then, the neuron number of the output layer of SOM is set as a correction factor to build the smoothing factor correction function in the form of the exponential function. The SOMPNN classifier is trained with the corrected smoothing factor. Experimental results demonstrate that the performance of the Haar-NMF+ISOMPNN-based vehicle detection classifier is improved in the detection rate, false detection rate and detection time compared with the traditional Haar+SOMPNN-based algorithm. © 2016, Editorial Department of Journal of Southeast University. All right reserved.