Pedestrian detection via multi-scale feature fusion convolutional neural network

Guo, Aixin; Yin, Baocai; Zhang, Jing; Yao, Jinfa

doi:10.1109/cac.2017.8242979

Cited by 8 publications

(7 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Their method uses max-pooling at the lower convolutional layers and adds a deconvolution operation at the deeper convolutional layers for up-sampling. Guo et al proposed a multi-scale feature fusion convolutional neural network (MFF-CNN) [23] for pedestrian detection. Similar to ION, features from the latter three convolutional layers are pooled and then spliced and fused.…”

Section: B Multi-scale Feature Fusion Methodsmentioning

confidence: 99%

An Improved Faster R-CNN Pedestrian Detection Algorithm Based on Feature Fusion and Context Analysis

et al. 2020

View full text Add to dashboard Cite

Considering the multi-scale and occlusion problem of pedestrian detection in natural scenes, we propose an improved Faster R-CNN pedestrian detection algorithm based on feature fusion and context analysis (FCF R-CNN). We design a feature fusion method of progressive cascade on VGG16 network, and add LRN to speed up the convergence of the network. The improved feature extraction network enables our model to generate high-resolution feature maps containing rich, detailed and semantic information. We also adjust the RPN parameters to improve the proposal efficiency. In addition, we add a multi-layer iterative LSTM module to the detection model, which uses LSTM's memory ability to extract the global context information of the candidate boxes. This method only needs the feature map of the image itself as input, which highlights useful context information and enables the model to generate more accurate candidate boxes containing potential pedestrians. Our method performs better than existing methods in detecting small-size and occluded pedestrians, and has strong robustness in challenging scenes. Our method achieves competitive results in both accuracy and speed on Caltech pedestrian dataset, achieving a LAMR value of 36.75% and a runtime of 0.20 seconds per image. The validity of the algorithm has been proved.INDEX TERMS context analysis, Faster R-CNN, feature fusion, pedestrian detection.

show abstract

Section: B Multi-scale Feature Fusion Methodsmentioning

confidence: 99%

An Improved Faster R-CNN Pedestrian Detection Algorithm Based on Feature Fusion and Context Analysis

et al. 2020

View full text Add to dashboard Cite

show abstract

“…[11] proposed a MOD method based on the Faster R-CNN boxwork, where a feature fusion module is introduced to complement the fine-grained knowledge of small-scale objects in the final features by fusing the strong semantic features in the top layer into the fine-resolution intermediate layer. [12] proposed a multi-scale feature fusion convolutional neural network MFF-CNN for pedestrian detection. It is similar to ION and also pools the features output from the last three convolutional layers and then splices and fuses them, only adding unique scaling and normalization methods for pedestrian features.…”

Section: Related Workmentioning

confidence: 99%

Multi-scale feature fusion object detection with bidirectional feature pyramid network and secondary candidate box refinement

Sun

Liu

et al. 2023

Preprint

View full text Add to dashboard Cite

As a task to locate and classify objects of interest in given images or videos, object detection is a core research direction of computer vision. It promotes intelligent monitoring, face recognition, and image segmentation development. However, deep learning-based object detection models still have a low detection accuracy for small-scale objects. In this paper, we propose a multi-scale feature fusion object detection with a bidirectional feature pyramid network and 1 secondary candidate box refinement, named MS-Faster R-CNN. The strategy, combined with the Feature Pyramid Network(FPN), uses two links to complete the feature fusion, making the semantics of the fused features richer and adapted to different scale objects. In the candidate box recommendation stage, we use the cascaded Region Proposal Network(RPN) and optimized Non-Maximum Sup-pression(NMS) so that the candidate box of the small-scale object would not be over-suppressed, which improves the recommendation efficiency of the candidate box. Finally, the Region of Interest(ROI) Align pooling technology based on the bilinear interpolation method is considered to avoid the loss of accuracy caused by quantization. Extensive experiments show that our scheme has better detection performance than other methods on datasets of Pascal VOC 2007, Pascal VOC 2012, and MS COCO.

show abstract

“…Afterwards, feature maps are merged again, which represents the actual fusion. MFF-CNN approaches have been used for different approaches like face recognition, [21], [22] and pedestrian detection, [23] and showed high performance. Moreover, GoogLeNet [24] and ResNet [25] can be defined as MFF-CNNs, due to their inception module or building block architecture, respectively.…”

Section: Fusion In Convolutional Neural Networkmentioning

confidence: 99%

Deep Metadata Fusion for Traffic Light to Lane Assignment

Langenberg

Lüddecke

Wörgötter

2019

IEEE Robot. Autom. Lett.

View full text Add to dashboard Cite

We present a deep metadata fusion approach that connects image data and heterogeneous metadata inside a Convolutional Neural Network (CNN). This approach enables us to assign all relevant traffic lights to their associated lanes. To achieve this, a common CNN topology is trained by down-sampled and transformed input images to predict an indication vector. The indication vector contains the column positions of all the relevant traffic lights that are associated with lanes. In parallel, we fuse prepared and adaptively weighted Metadata Feature Maps (MFM) with the convolutional feature map input of a selected convolutional layer. The results are compared to rule-based, only-metadata, and only-vision approaches. In addition, human performance of the traffic light to ego-vehicle lane assignment has been measured by a subjective test. The proposed approach outperforms all other approaches. It achieves about 93.0% average precision for a real-world dataset. In a more complex dataset, 87.1% average precision is achieved. In particular, the new approach reaches significantly higher results with 93.7% to 91.0% average accuracy for a real-world dataset in contrast to lower human performance.

show abstract

Pedestrian detection via multi-scale feature fusion convolutional neural network

Cited by 8 publications

References 19 publications

An Improved Faster R-CNN Pedestrian Detection Algorithm Based on Feature Fusion and Context Analysis

An Improved Faster R-CNN Pedestrian Detection Algorithm Based on Feature Fusion and Context Analysis

Multi-scale feature fusion object detection with bidirectional feature pyramid network and secondary candidate box refinement

Deep Metadata Fusion for Traffic Light to Lane Assignment

Contact Info

Product

Resources

About