Year: 2011
Type: article
Abstract: [EN] In this paper, we introduce a hybrid real-time \nmethod for vision-based pedestrian detection \nmade up by the sequential combination of two \nbasic methods applied in a coarse to fine fashion. \nThe proposed method aims to achieve an \nimproved balance between detection accuracy \nand computational load by taking advantage \nof the strengths of these basic techniques. \nHaar-like features combined with Boosting \ntechniques, which have been demonstrated to \nprovide rapid but not accurate enough results in \nhuman detection, are used in the first stage to \nprovide a preliminary candidate selection in the \nscene. Then, feature extraction and classification \nmethods, which present high accuracy rates at \nexpenses of a higher computational cost, are \napplied over boosting candidates providing the \nfinal prediction. Experimental results show that \nthe proposed method performs effectively and \nefficiently, which supports its suitability for real \napplications.
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2182613072