On-vehicle pedestrian area estimation method based on optical flow clustering

Abstract

The present invention discloses an on-vehicle pedestrian area estimation method based on optical flow clustering, which comprises: carrying out optical flow estimation on the image obtained by a camera to obtain an image optical flow field; performing optical flow clustering again so as to estimate the background area; eliminating the background area; using the graph segmentation algorithm to segment the foreground area; finally, discriminating the areas in the foreground area; and identifying the valid pedestrian area. The on-vehicle pedestrian area estimation method based on optical flow clustering proposed by the invention avoids the blind search for pedestrian recognition caused by global template searching and matching adopted by a traditional pedestrian detection system, which is suitable for intelligently assisted driving or unmanned driving of vehicles in an on-vehicle environment and overcomes the problem that it is difficult to determine the Gaussian distribution of the background by common EM algorithm and clustering algorithm such as K-mean algorithm. The optical flow clustering algorithm proposed by this method can estimate the background area more effectively. In the process of discriminating the foreground pedestrian areas, the non-pedestrian area can be effectively eliminated and the pedestrian area can be obtained by using the morphological characteristics of the human body.

Description

technical field [0001] The invention belongs to the field of intelligent transportation and intelligent vehicle research, and relates to a vehicle-mounted pedestrian area estimation method based on optical flow clustering, which is suitable for quickly locating pedestrian areas in a pedestrian detection system. Background technique [0002] In recent years, research results such as Haar wavelet pedestrian detection method, human body shape detection method, HOG+SVM pedestrian detection method have been gradually formed, which have promoted the development of intelligent driving technology. These studies have adopted the pedestrian template matching method of sliding window to identify pedestrians. However, due to the global multi-scale window search strategy of template matching, a large number of blind searches and time waste are caused in pedestrian detection, which reduces the real-time performance of pedestrian detection. The method of obtaining the possible pedestrian a...

AI Technical Summary

Problems solved by technology

Although this type of method can obtain pedestrian ROI, the complex model calculation increases the running time of pedestrian detection. (2) Motion detection method, which uses optical flow calculation and motion segmentation method to obtain pedestrian area

For example, Elzein et al. provided a new vision for pedestrian ROI estimation from the perspective of motion segmentation, but this method lacks consideration of the motion difference characteristics of motion scenes and pedestrians, and still cannot reduce the computational complexity of optical flow field segmentation.

Images