An Improved Pedestrian Detection Algorithm using Integration of Resnet and Yolo V2

Abstract

Pedestrian detection is one of the important tasks in object detection technology. The pedestrian detection algorithm has been used in applications like intelligent video surveillance, traffic analysis, and autonomous driving. In recent years, many pedestrian detection algorithms have been proposed but the key drawback is the accuracy and speed, which can be improved my integrating efficient algorithms. The proposed model improves the pedestrian detection algorithm by integrating two efficient algorithms together. The model is developed using the joint version of ResNet and YOLO v2, which preforms feature extraction and classification respectively. By using this model the efficiency of the system is increased by improving the accuracy rate so that it can be used with real-time applications. The model has been compared with existing models like SSD, Faster R-CNN and Mask R-CNN. Comparing with these models, the proposed model provides mAP value higher than these existing models with less loss function when tested on the INRIA dataset.

Research on Intelligent Embedded Fan System Based on YOLOv2 Lightweight Algorithm

“…erefore, YOLOv2 model algorithm is selected. e YOLOv2 network can be directly loaded through the KPU module [17] and then combined with the face detection model to realize face recognition [18][19][20].…”

“…Although YOLOv1 detection speed is fast, YOLOv1 is inferior to R-CNN, which is not accurate in object positioning and has low recall. YOLOv2 abandons dropout, convolves and adds batch normalization, normalizes each layer of the network, improves the convergence speed, performs batch normalization processing, and improves the mAP [18].…”