Fast Human Detection Based On The Census Transform Histogram

Human detection can detect human in video and image and analyse the related dataautomatically. It is an attractive research subject in computer vision. It is important in a widerange of applications, such as intelligent monitoring system, Intelligent Transportation,motion analysis and man-machine interface.This paper first introduces the background and difficulties of the human detectiontechnology, then gives a comprehensive overview of the existing algorithms. Recent progressin human detection has advanced in many aspects, e.g., features, classifiers, testing speed, andocclusion handing. However, the real time detection and detection accuracy is still a problem.By analyzing the shortcomings of the existing algorithms, we propose a fast algorithm basedon the Census Transform histogram algorithm and apply it in human detection．This algorithm includes video frame image extraction, image pretreatment, the CensusTransform histogram feature vector extraction and Classification. Consider of the relevance ofvideo images, catching a key frame from every two images. To obtain the inte rested regionsbased on the conterminous Frame difference and motion history image. Then enhance theimage using Sobel operators. Through the signs of the comparisons among neighboring pixels,it will get the histogram of the detection window. We can encode the global contour by thedistribution coefficient of the histogram. When the Linear SVM is used, it will be classified.Considering of the longer processing time in the existing algorithm, we propose a fastcomputational method that does not need to explicitly generate feature vectors and not requirefeature vectors normalization. This method has higher efficiency and can’t reduce theaccuracy.This paper has a library which contains a training set with3400images and a test setwith1500pictures. Experiment shows that the method based on the Census Transformhistogram algorithm is reliable and effective. It achieves25fps speed with83.5%recognitionrate and can be used in a real-time system.