Ten Years of Pedestrian Detection, What Have We Learned?

By Rodrigo Benenson, Mohamed Omran, Jan Hendrik Hosang, Bernt Schiele. ECCV Workshops, 2014

724 citations recorded by the source corpus.

Paper-by-paper results make it easy to miss the forest for the trees.We analyse the remarkable progress of the last decade by dis- cussing the main ideas explored in the 40+ detectors currently present in the Caltech pedestrian detection benchmark. We observe that there exist three families of approaches, all currently reaching similar detec- tion quality. Based on our analysis, we study the complementarity of the most promising ideas by combining multiple published strategies. This new decision forest detector achieves the current best known performance on the challenging Caltech-USA dataset.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/8bedc96cafc9ff4771464a635655a3bdb5cefce0

How researchers see this paper

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

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Occluded Pedestrian Detection Through Guided Attention in CNNs (2018, background): “During the last decade, great progress has been achieved [3, 31], especially by properly adapting CNNs for general object detection to this canonical task [12, 30, 32].”
DecideNet: Counting Varying Density Crowds Through Attention Guided Detection and Density Estimation (2017, methodology): “From Table 1, we can observe the detection based approaches [1, 7, 25] generally perform worse than the regression counterparts.”
Multispectral Pedestrian Detection using Deep Fusion Convolutional Neural Networks (2016, background): “Although the topic has been intensively investigated in the last decade [1, 2, 3], it is still a challenging task.”
How Far are We from Solving Pedestrian Detection? (2016, background): “SCF [3] 35 5pp 4pp Checkerboards 19 ~0 1pp”
Illumination-aware Faster R-CNN for Robust Multispectral Pedestrian Detection (2018, background): “Pedestrian detection, as a canonical sub-problem of general object detection, has been intensively investigated by the computer vision community [1, 2, 3], due to its diversiﬁed applications in video surveillance, car safety, image retrieval, robotics, etc.”
Computer vision and deep learning techniques for pedestrian detection and tracking: A survey (2018, background): “Since occlusions are one of the most discussed problems in literature [25,26,6 8,82,93,14 9,193] , deep learning allowed to strengthen the detection of single parts in order to ﬁnd and correctly classify occluded pedestrian [82,88,115,116] .”
Taking a deeper look at pedestrians (2015, methodology): “Caltech The Caltech dataset and its associated benchmark [13, 5] is one of the most popular pedestrian detection datasets.”
EuroCity Persons: A Novel Benchmark for Person Detection in Traffic Scenes (2018, methodology): “In [12] more than 40 detectors are evaluated on the Caltech dataset to analyze the main cause for improvement during the last 10 years.”
Perceptual Generative Adversarial Networks for Small Object Detection (2017, methodology): “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] !" #$%%# &'(”
Performance Measures and a Data Set for Multi-target, Multi-camera Tracking (2016, background): “MTMC trackers rely on pedestrian detection [15] and tracking [16] or assume single-camera trajectories to be given [6, 13, 17–26].”
CityPersons: A Diverse Dataset for Pedestrian Detection (2017, background): “Initial attempts to apply convnets for pedestrian detection, used existing detectors (mainly decision forests over hand-crafted features [2, 33]) outputs and re-scored them with a convnet classifier (plus bounding box regression) [16, 28, 1, 27].”
In Teacher We Trust: Learning Compressed Models for Pedestrian Detection (2016, methodology): “Teacher Network For our teacher network, We use pre-activation ResNet-200 [16] pre-trained on ImageNet, augmented with dropout and a 64-dimensional hint layer, then ﬁne-tuned on our training set.”
Filtered channel features for pedestrian detection (2015, methodology): “The most recent survey [2] indicates that the classifier choice (e.g. linear/non-linear SVM versus decision forest) is not a clear differentiator regarding quality; rather the features used seem more important.”