Research On Algorithm Of Pedestrian Detection Integrating With Depth Information

Pedestrian detection is technique of utilizing the computer vision to judge the presence of pedestrians in images or video sequences,and gives precise positioning of pedestrians.In the field of intelligent monitoring,intelligent transportation,human-computer interaction and other fields,pedestrian detection have a broad application prospects.Because pedestrians are not rigid,their clothes are different,and their background is complex and changeable,those challenges have brought great difficulties to pedestrian detection.Although the color image used in pedestrian detection has rich texture details,high resolution,and can provide ample information available,they are easily affected by factors such as illumination,occlusion,and complex background.With the advent of the Microsoft Kinect camera,use the depth image to detect person has become a new research hotspot.Although the resolution of the depth image is limited,it is invariant to the change of illumination and can complement the advantages of the color image.In order to improve the accuracy and robustness of pedestrian detection algorithm,this paper combines the information of color images and depth images from the perspective of pixel-level and feature-level information fusion.The main contributions are as follows:1.The algorithm of pedestrian detection based on feature of HOG from RGB-D fusion image is proposed.The algorithm firstly uses the Haar wavelet to do pixel level fusion for the color image and the depth image,and then extracts the gradient direction histogram feature of the fused image.Finally,the feature descriptor is used to classify the pedestrians,which can effectively cope with illumination,occlusion,complex backgrounds and other challenges.Experimental results show that compared with other classical algorithms,the detection accuracy of this algorithm has been significantly improved.2.Proposes a pedestrian detection algorithm that integrate the edge of the color image and depth direction histogram feature.The algorithm firstlydescribes the pedestrian's overall structural properties by extracting the edge features from the color image through the shearlet transform,and extract the depth gradient histogram from the depth image to obtain the local region's gradient and edge information,and then fuse them as a new feature descriptor to detect pedestrian.Due to the full integration of the pedestrian's local feature and overall structural feature,our algorithm has obvious advantages when facing with challenges such as noise,occlusion,lighting,and similar colors.Experiments show that compared with other algorithms,the robustness of our algorithm has been significantly improved.