Fast Moving Human Detection Using Fourier and HOG Descriptors Based CUDA

The most sensitive areas have integrated video surveillance systems capable of detecting essentially the suspicious people. These systems consist of detecting human among other moving objects. Therefore, accuracy and speed of video surveillance systems are necessary more than ever. However, a compromise must be found between precision and real-time detection. In one hand, to obtain an excellent precision, the detection task should be subdivided in order to treat smallest details thus consuming more time. In the other hand, minimizing the processing time results in too much false detection. In our work, we describe a new human detection approach based on Fourier and Histogram of Orient Gradient descriptors using CUDA the parallel architecture of GPU. As SVM is able to provide the posterior probability, we use this binary classifier to integrate both types of feature descriptors to achieve best performances. Experimental results show that the proposed approach is efficient and outperforms some existing techniques.