Filtered channel features for pedestrian detection

By Shanshan Zhang, Rodrigo Benenson, Bernt Schiele. 2015

387 citations recorded by the source corpus.

This paper starts from the observation that multiple top performing pedestrian detectors can be modelled by using an intermediate layer filtering low-level features in combination with a boosted decision forest. Based on this observation we propose a unifying framework and experimentally explore different filter families. We report extensive results enabling a systematic analysis. Using filtered channel features we obtain top performance on the challenging Caltech and KITTI datasets, while using only HOG+LUV as low-level features. When adding optical flow features we further improve detection quality and report the best known results on the Caltech dataset, reaching 93% recall at 1 FPPI.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

End-to-End People Detection in Crowded Scenes (2015, methodology): “Our approach achieves EER of 80% compared to 70% for [27].”
A Unified Multi-scale Deep Convolutional Neural Network for Fast Object Detection (2016, methodology): “The model “h720-ctx” was compared to methods such as DeepParts [39], CompACT-Deep [15], CheckerBoard [40], LDCF [43], ACF [2], and SpatialPooling [41] on three tasks: reasonable, medium and partial occlusion.”
Pedestrian detection aided by deep learning semantic tasks (2014, background): “This problem is challenging because of large variation and confusion in human body and background scene, as shown in Fig.1 (a), where the positive and hard negative patches have large ambiguity.”
Learning Complexity-Aware Cascades for Deep Pedestrian Detection (2015, background): “Several works proposed alternative feature channels, obtained by convolving different filters with the original HOG+LUV channels [36, 37, 1, 21].”
How Far are We from Solving Pedestrian Detection? (2016, methodology): “Figure 4: Errors analysis of Checkerboards [23] on the test set.”
End-to-End Deep Learning for Person Search (2016, background): “ods with scores re-weighting to solve the problem, while in this work we propose a deep learning framework that jointly handles both aspects. Pedestrian detection. DPM [7], ACF [6], and Checkerboards [41] are the most commonly used off-the-shelf pedestrian detectors. They rely on hand-crafted featur”
Perceptual Generative Adversarial Networks for Small Object Detection (2017, neutral): “We compare the result of Perceptual GAN with all the existing methods that achieved best performance on the Caltech testing set, including VJ [37], HOG [5], LDCF [27], Katamari [2], SpatialPooling+ [30], TA-CNN [36], Checkerboards [43], CompACT-Deep [44] !" #$%%# &'(”
What Can Help Pedestrian Detection? (2017, methodology): “, human behavior analysis, gait recognition, intelligent video surveillance and automatic driving, has attracted massive attention in the last decade [11, 26, 10, 35, 33, 30].”
KGSNet: Key-Point-Guided Super-Resolution Network for Pedestrian Detection in the Wild (2020, methodology): “…paradigm, designing effective hand-crafted features (e.g., Haar [5] and HOG [6], [7]), training a stronger classiﬁer (e.g., SVM [6] and boosting classiﬁers [8]) and combining features from multiple channels (e.g., [9] and [10]) are of critical importance for achieving state-of-the-art performance.”
3D Object Proposals Using Stereo Imagery for Accurate Object Class Detection (2016, neutral): “38 FilteredICF [59] - 67.”
Subcategory-Aware Convolutional Neural Networks for Object Proposals and Detection (2016, neutral): “urs) 90.74 88.55 77.95 90.49 87.88 77.10 Pedestrian ACF [23] 44.49 39.81 37.21 N/A N/A N/A DPM [12] 47.74 39.36 35.95 43.58 35.49 32.42 DPM-VOC+VP [28] 59.48 44.86 40.37 53.55 39.83 35.73 FilteredICF [33] 67.65 56.75 51.12 N/A N/A N/A DeepParts [34] 70.49 58.67 52.78 N/A N/A N/A Regionlets [30] 73.1”
CrowdHuman: A Benchmark for Detecting Human in a Crowd (2018, methodology): “Detecting people in images is among the most important components of computer vision and has attracted increasing attention in recent years [29, 14, 32, 30, 10, 5, 4, 6, 18].”
Scale-Aware Fast R-CNN for Pedestrian Detection (2015, methodology): “The detection results and performance comparisons of the proposed method with several best-performing methods [17] [25] [44] [34] [46] [41] [4] are presented in Table I and Figure 7.”
Is Faster R-CNN Doing Well for Pedestrian Detection? (2016, methodology): “3 shows that in general RPN performs better than three leading methods that are based on traditional features: SCF [11], LDCF [24] and Checkerboards [31].”
CityPersons: A Diverse Dataset for Pedestrian Detection (2017, background): “ACF [8] and Checkerboards [33] are representatives from the Integral Channel Features detector (ICF) family, while FasterRCNN [26] acts as the state-ofthe-art detector.”
Occlusion-aware R-CNN: Detecting Pedestrians in a Crowd (2018, background): “[66] provide a systematic analysis for the filtered channel features, and find that with the proper filter bank, filtered channel features can reach top detection quality.”
Taking a deeper look at pedestrians (2015, background): “While preparing this manuscript, additional related papers appeared [46, 54].”