Fused DNN: A Deep Neural Network Fusion Approach to Fast and Robust Pedestrian Detection

Abstract

We propose a deep neural network fusion architecture for fast and robust pedestrian detection. The proposed network fusion architecture allows for parallel processing of multiple networks for speed. A single shot deep convolutional network is trained as a object detector to generate all possible pedestrian candidates of different sizes and occlusions. This network outputs a large variety of pedestrian candidates to cover the majority of ground-truth pedestrians while also introducing a large number of false positives. Next, multiple deep neural networks are used in parallel for further refinement of these pedestrian candidates. We introduce a soft-rejection based network fusion method to fuse the soft metrics from all networks together to generate the final confidence scores. Our method performs better than existing state-of-the-arts, especially when detecting small-size and occluded pedestrians. Furthermore, we propose a method for integrating pixel-wise semantic segmentation network into the network fusion architecture as a reinforcement to the pedestrian detector. The approach outperforms stateof-the-art methods on most protocols on Caltech Pedestrian dataset, with significant boosts on several protocols. It is also faster than all other methods.

Improved Hough transform by modeling context with conditional random fields for partially occluded pedestrian detection

“…[21][22][23][24][25][26][27][28][29][30][31][32][33][34][35][36][37][38][39] Here, we review the methods based on the Hough transform framework 1,2,4-68-10 that are most relevant to our work.…”

Hybrid channel based pedestrian detection

“…F-DNN [34] uses a soft-rejection to fuse multiple deep neural networks to classify the candidates. F-DNN+SS [34] further uses a pixel-wise semantic segmentation network to refine the classification and improves accuracy at the expense of a significant loss in speed. GDFL [33] includes three components: a convolutional back-bone, a scale-aware pedestrian attention module and a zoom-inzoom-out module to identify small and occluded pedestrians.…”

Monitoring the Number of Pedestrians in an Area: The Applicability of Counting Systems for Density State Estimation

“…The detection approaches vary widely from detection using global models [21][22][23] to detection based on more local features and/or parts of pedestrians [24][25][26]. More and more, machine learning strategies are used to improve the detection of pedestrians under a multitude of conditions (e.g., [27,28]). Also a multitude of tracking algorithms have been presented, some of which use intermediate models to predict the future movement of pedestrians (e.g., [29]), while others use the information of the image to determine the motion between images (e.g., [30,31]).…”