Fast pedestrian detection based on region of interest and multi-block local binary pattern descriptors

• We introduced a new ROI method, it decreases computational cost and improve detection speed.
• We improved the images edges within the candidate region using non-tensor product wavelet filter.
• We used full-body descriptor based on shapelet and body-part descriptor based on MB-LBP for features extraction.
• We built a cascade ILFDA classifier to handle large data.

Nowadays pedestrian detection plays a crucial role in image or video retrieval, video monitoring systems and driving assistance systems. Detecting moving pedestrians is a challenging task, some of the detection methods are ineffective and slow. Occlusion, rotation, changes in object shapes, real time detection and illumination conditions are predominant obstacles. This paper is focus on the implementation of an efficient and speedy detector. A detection framework based on region of interest (ROI), full-body descriptor, body-part descriptors, and cascade classifier is proposed. ROI identifies, locates, and extracts candidate regions containing pedestrians, thus reducing the number of detection windows. In relation to human detection, independent information sources such as shapelet features and multi-block local binary pattern (MB-LBP) are used for features extraction. Experimental results showed that the proposed-model performs better than some state-of-the-art approaches, with suitable processing time for further operations such as tracking and imminent danger estimation.

Figure optionsDownload as PowerPoint slide