Combining ROI-base and Superpixel Segmentation for Pedestrian Detection

Pedestrian Detection is a hot topic in recent years, which is attracting a large number of scholars. The detection models are developing from simple models to complex models and the detection accuracy has been greatly improved. DPM (deformable part model) become the best pedestrian detection model and also attracted many scholars to modify it. The biggest problem caused by complex models is low detection efficiency for the real-time application with the sliding windows framework. Meanwhile, the latent SVM algorithm in DPM mining parts information is greatly affected by the initialization of parts, and there is no exact solution. Aiming at the drawbacks of DPM, using the research achievement of salient object detection an background detection, we propose a novel pedestrian detection framework based on ROI and superpixel segmentation. Contrasting with DPM in experiments, our method have greatly improved in accuracy and efficiency. The proposed framework has the same reference to other complex models.