Pedestrian Detection: Performance Comparison Using Multiple Convolutional Neural Networks

Pedestrian Detection in real world crowded areas is still one of the challenging categories in object detection problems. Various modern detection architectures such as Faster R-CNN, R-FCN and SSD has been analyzed based on speed and accuracy measurements. These models can detect multiple objects with overlaps and localize them using a bounding box framing it. Evaluation of performance parameters provides high speed models which can work on live stream applications in mobile devices or high accurate models which provide state-of-the-art performance for various detection problems. These convolutional neural network models are tested on the Penn-Fudan Dataset as well as Google images with occlusions, which achieves high detection accuracies on each of the detectors.