The Fastest Pedestrian Detector in the West

By Piotr Doll'ar, Serge J. Belongie, Pietro Perona. 2010

682 citations recorded by the source corpus.

We demonstrate a multiscale pedestrian detector operating in near real time ( 6 fps on 640x480 images) with state-of-the-art detection performance. The computational bottleneck of many modern detectors is the construction of an image pyramid, typically sampled at 8-16 scales per octave, and associated feature computations at each scale. We propose a technique to avoid constructing such a finely sampled image pyramid without sacrificing performance: our key insight is that for a broad family of features, including gradient histograms, the feature responses computed at a single scale can be used to approximate feature responses at nearby scales. The approximation is accurate within an entire scale octave. This allows us to decouple the sampling of the image pyramid from the sampling of detection scales. Overall, our approximation yields a speedup of 10-100 times over competing methods with only a minor loss in detection accuracy of about 1-2% on the Caltech Pedestrian dataset across a wide range of evaluation settings. The results are confirmed on three additional datasets (INRIA, ETH, and TUD-Brussels) where our method always scores within a few percent of the state-of-the-art while being 1-2 orders of magnitude faster. The approach is general and should be widely applicable.

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

Pedestrian Detection with Unsupervised Multi-stage Feature Learning (2012, background): “While processing speed is not the focus of this paper, features and classifier approximations introduced by [8] and [3] may be applicable to deep learning models for faster detection, in addition to GPU optimizations.”
Fast Edge Detection Using Structured Forests (2014, background): “and gradient channels (originally developed for fast pedestrian detection [12]).”
A ConvNet for the 2020s (2022, background): “For many decades, this has been the default use of ConvNets, generally on limited object categories such as digits [43], faces [58, 76] and pedestrians [19, 63].”
Deep Feature Pyramid Reconfiguration for Object Detection (2018, background): “To make them scale-invariant, these features are computed over image pyramids [9,13].”
Pedestrian Detection: An Evaluation of the State of the Art (2012, methodology): “Moreover, numerous approaches have been proposed for speeding up detection, including speeding up the detector itself [29], [44], [46], through use of approximations [63], [89] or by using special purpose hardware such as GPUs [90] (for a review of fast detection see [63]).”
Exploring Weak Stabilization for Motion Feature Extraction (2013, background): “Walk et al. [28] proposed a number of modifications to the HOF features that resulted in mod- est gains in detection performance.”
Structured Forests for Fast Edge Detection (2013, methodology): “[21], we use a similar set of color and gradient channels (originally developed for fast pedestrian detection [8]).”
Extended Object Tracking: Introduction, Overview and Applications (2016, methodology): “For each image in the dataset a pedestrian detector [40], [41] is used, and the measurements are projected onto the ground plane using the camera parameters.”
Person Re-identification Using the Silhouette Shape Described by a Point Distribution Model (2015, methodology): “We chose to implement the Fastest Pedestrian Detector in the West [6] which has the advantage of working in a single-shot mode.”
AutoFocus: Efficient Multi-Scale Inference (2018, methodology): “Common techniques involved approximation of features to reduce the number of scales [17, 3], cascades [6, 16] or feature pyramids [15].”
Pedestrian detection at 100 frames per second (2012, methodology): “This detector is 20 times faster than previous results of equal quality [6, 8], and has only half as many false positives on the INRIA dataset compared to previous results obtained at equal speed [17].”
Robust Multi-resolution Pedestrian Detection in Traffic Scenes (2013, methodology): “Traditional pedestrian detectors usually follow the scale invariant assumption: a scale invariant feature based detector trained at a fixed resolution could be generalized to all resolutions, by resizing the detector [40, 4], image [6, 19] or both of them [11].”
A discriminative deep model for pedestrian detection with occlusion handling (2012, background): “Detectors using boosting to select features from a large pool of local candidate features also consider objects as being composed of parts [6, 7, 25].”
Fast Feature Pyramids for Object Detection (2014, methodology): “ACF uses the same channels as [39]: normalized gradient magnitude, histogram of oriented gradients (six channels), and LUV color channels.”
Learning Optimal Embedded Cascades (2012, background): “According to [32], the best current pedestrian detection results are due to [33].”
Single-Pedestrian Detection Aided by Multi-pedestrian Detection (2013, methodology): “The evaluated 1-pedestrian detectors are VJ [32], Shapelet [27], PoseInv [20], LatSVM-V1 [13], HikSVM [21], HOG [4], MultiFtr [35], HogLbp [34], Pls [28], MultiFtr+CCS, MultiFtr+Motion [33], FPDW [8], ChnFtrs [9], and MultiResC [25].”
Local Transform Features and Hybridization for Accurate Face and Human Detection (2013, background): “Index Terms—Local binary pattern, local gradient pattern, binary histograms of oriented gradients, feature hybridization, face and human detection Ç”