#### DMCA

## Traffic sign recognition - How far are we from the solution? IJCNN (2013)

Citations: | 13 - 1 self |

### Citations

6478 | Libsvm: a library for support vector machines
- Chang, Lin
- 2011
(Show Context)
Citation Context ... Kernel- min(x,y), Polynomial - (x·y+1)5, and Radial Basis Function exp(−‖x−y‖2). The classifiers are named LSVM, IKSVM, POLYSVM, and RBFSVM, accordingly. We train one-vs-all classifiers using LIBSVM =-=[24]-=- (with parameter C = 10) and LIBLINEAR [25] (with parameters C = 10, B = 10). As in [15], the test sample is associated with the class with the highest posterior probability estimated using the sample... |

3731 | Histograms of Oriented Gradients for Human Detection
- Dalal, Triggs
- 2005
(Show Context)
Citation Context ...introduced by Dollar et al. [11], builds upon these features to provide excellent quality for the task of pedestrian detection, significantly improving over the traditional HOG+linear SVM combination =-=[12]-=-. More recently, variants of the integral channel features detector have shown to reach high speed (50 fps) [13], [14], and top quality, improving over most existing methods for pedestrian detection [... |

3688 | Support-vector networks
- Cortes, Vapnik
- 1995
(Show Context)
Citation Context ...lass and assigns the query to the class with the maximum such sum. 4) Support Vector Machines Classifiers: Support Vector Machines (SVM) are a very popular technique for out of the box classification =-=[22]-=-, [23]. The kernels k(x,y) we use here are the Linear - x·y, Intersection Kernel- min(x,y), Polynomial - (x·y+1)5, and Radial Basis Function exp(−‖x−y‖2). The classifiers are named LSVM, IKSVM, POLYSV... |

1413 | LIBLINEAR: A library for large linear classification
- Fan, Chang, et al.
(Show Context)
Citation Context ...nd Radial Basis Function exp(−‖x−y‖2). The classifiers are named LSVM, IKSVM, POLYSVM, and RBFSVM, accordingly. We train one-vs-all classifiers using LIBSVM [24] (with parameter C = 10) and LIBLINEAR =-=[25]-=- (with parameters C = 10, B = 10). As in [15], the test sample is associated with the class with the highest posterior probability estimated using the sample’s margin. VI. CLASSIFICATION EXPERIMENTS I... |

933 | Robust face recognition via sparse representation
- Wright, Yang, et al.
- 2009
(Show Context)
Citation Context ...s given in [5]. A. Contributions The contribution of this work is three-fold: • We show that top performance can be reached using existing approaches for pedestrian detection [6] and face recognition =-=[7]-=-, [8], without the need to encode traffic sign specific information. • We provide an extensive evaluation on two large Belgian and German traffic sign benchmarks for both detection and classification.... |

664 | Laplacian eigenmaps and spectral techniques for embedding and clustering
- Belkin, Niyogi
- 2001
(Show Context)
Citation Context ...ses. 2) Sparse Representation based Linear Projection (SRLP) [10]: is a variant of Locality Preserving Projections (LPP) [18]. LPP itself is a linear approximation of the nonlinear Laplacian Eigenmap =-=[19]-=- method aiming at preserving the local affinities from the original space into the embedding. The algorithmic procedure employs an adjacency graph construction, setting edge weights, and finally solvi... |

469 | PCA versus LDA
- Martinez, Kak
- 2001
(Show Context)
Citation Context ...sification. In the following we shortly review the dimensionality reduction techniques used. The reader is referred to the original works for more details. 1) Linear Discriminant Analysis (LDA) [16], =-=[17]-=-: is an embedding technique, which maximizes the inter-class variance while minimizing the intra-class variance. The LDA projection thus tries to best discriminate among classes. The solution can be o... |

441 | Efficient sparse coding algorithms
- Lee, Battle, et al.
(Show Context)
Citation Context ...osition of a query sample over a set of training samples. This, in practice, is obtained solving an l1-regularized least squares formulation. In all our experiments, we use the Feature Sign algorithm =-=[21]-=- and fix its regulatory parameter to 0.05. The Sparse Representation-based Classifier (SRC) [7] decision is taken based on the corresponding residuals to each class training samples. For speed, we dir... |

414 | Locality preserving projections.
- He, Niyogi
- 2003
(Show Context)
Citation Context ... to an embedding space with a number of dimensions less than the number of classes. 2) Sparse Representation based Linear Projection (SRLP) [10]: is a variant of Locality Preserving Projections (LPP) =-=[18]-=-. LPP itself is a linear approximation of the nonlinear Laplacian Eigenmap [19] method aiming at preserving the local affinities from the original space into the embedding. The algorithmic procedure e... |

254 | Classification using intersection kernel support vector machines is efficient
- Maji, Berg, et al.
(Show Context)
Citation Context ...nd assigns the query to the class with the maximum such sum. 4) Support Vector Machines Classifiers: Support Vector Machines (SVM) are a very popular technique for out of the box classification [22], =-=[23]-=-. The kernels k(x,y) we use here are the Linear - x·y, Intersection Kernel- min(x,y), Polynomial - (x·y+1)5, and Radial Basis Function exp(−‖x−y‖2). The classifiers are named LSVM, IKSVM, POLYSVM, and... |

133 |
The Statistical Utilization of Multiple Measurements
- Fisher
- 1938
(Show Context)
Citation Context ...e classification. In the following we shortly review the dimensionality reduction techniques used. The reader is referred to the original works for more details. 1) Linear Discriminant Analysis (LDA) =-=[16]-=-, [17]: is an embedding technique, which maximizes the inter-class variance while minimizing the intra-class variance. The LDA projection thus tries to best discriminate among classes. The solution ca... |

90 |
Integral channel features.
- Dollar, Tu, et al.
- 2009
(Show Context)
Citation Context ...tection. It is currently accepted that histogram of oriented gradients (HOG) is an effective way to capture shape information. The integral channel features detector first introduced by Dollar et al. =-=[11]-=-, builds upon these features to provide excellent quality for the task of pedestrian detection, significantly improving over the traditional HOG+linear SVM combination [12]. More recently, variants of... |

65 | Pedestrian detection at 100 frames per second
- Benenson, Mathias, et al.
- 2012
(Show Context)
Citation Context ...etection, significantly improving over the traditional HOG+linear SVM combination [12]. More recently, variants of the integral channel features detector have shown to reach high speed (50 fps) [13], =-=[14]-=-, and top quality, improving over most existing methods for pedestrian detection [6] (while still using a single rigid template per candidate detection window). In this paper we show we can use the in... |

41 | Crosstalk cascades for frame-rate pedestrian detection. ECCV,
- Dollar, Appel, et al.
- 2012
(Show Context)
Citation Context ...rian detection, significantly improving over the traditional HOG+linear SVM combination [12]. More recently, variants of the integral channel features detector have shown to reach high speed (50 fps) =-=[13]-=-, [14], and top quality, improving over most existing methods for pedestrian detection [6] (while still using a single rigid template per candidate detection window). In this paper we show we can use ... |

32 | An efficient algorithm for large-scale discriminant analysis".
- Cai, He, et al.
- 2008
(Show Context)
Citation Context ...wing combinations: LDA I, LDA PI, LDA HOGx : Linear Discriminant Analysis (LDA) projections of the original image feature representation (I, PI, HOG1, HOG2, or HOG3). Note that we use regularized LDA =-=[20]-=- (as in [10]). SRLP I, SRLP PI, SRLP HOGx : Sparse Representation-based Linear Projections (SRLP) of the original image feature representation. We use the regularized version of SRLP as introduced by ... |

30 | Traffic sign recognition with multi-scale convolutional networks.
- Sermanet, LeCun
- 2011
(Show Context)
Citation Context ...LASSIFICATION RESULTS ON GTSC CR (%) Team Method 99.46 IDSIA [26] Committee of CNNs 98.84 INI-RTCV [4] Human Performance 98.53 ours INNC+INNLP(I,PI,HOGs) 98.50 ours SRC+SRLP(I,PI,HOGs) 98.31 sermanet =-=[27]-=- Multi-scale CNNs 98.19 [28] SRGE+HOG2 96.14 CAOR [4] Random Forests 95.68 INI-RTCV [4] LDA on HOG2 The same effect is found on BTSC. Concatenating LDA projections of I and of PI features boosts the p... |

27 |
Towards fully autonomous driving: Systems and algorithms
- Levinson, Askeland, et al.
- 2011
(Show Context)
Citation Context ...ons for classification. I. INTRODUCTION TRAFFIC SIGN RECOGNITION (TSR) gets consider-able interest lately. The interest is driven by the market for intelligent applications such as autonomous driving =-=[1]-=-, advanced driver assistance systems (ADAS) [2], mobile mapping [3], and the recent releases of larger traffic signs datasets such as Belgian [3] or German [4] datasets. TSR covers two problems: traff... |

25 | Multi-view traffic sign detection, recognition
- Timofte, Zimmermann, et al.
- 2009
(Show Context)
Citation Context ...SR) gets consider-able interest lately. The interest is driven by the market for intelligent applications such as autonomous driving [1], advanced driver assistance systems (ADAS) [2], mobile mapping =-=[3]-=-, and the recent releases of larger traffic signs datasets such as Belgian [3] or German [4] datasets. TSR covers two problems: traffic sign detection (TSD) and traffic sign classification (TSC). TSD ... |

25 | Seeking the strongest rigid detector.
- Benenson, Mathias, et al.
- 2013
(Show Context)
Citation Context ...s and existing datasets is given in [5]. A. Contributions The contribution of this work is three-fold: • We show that top performance can be reached using existing approaches for pedestrian detection =-=[6]-=- and face recognition [7], [8], without the need to encode traffic sign specific information. • We provide an extensive evaluation on two large Belgian and German traffic sign benchmarks for both dete... |

19 | Multi-column deep neural network for traffic sign classification.
- D, Meier, et al.
- 2012
(Show Context)
Citation Context ...31 INNC 98.20 98.25 98.53 LSVM 97.82 97.47 97.25 IKSVM 97.78 98.15 98.13 POLYSVM 97.91 98.16 98.02 RBFSVM 97.94 98.16 98.05 TABLE IX BEST CLASSIFICATION RESULTS ON GTSC CR (%) Team Method 99.46 IDSIA =-=[26]-=- Committee of CNNs 98.84 INI-RTCV [4] Human Performance 98.53 ours INNC+INNLP(I,PI,HOGs) 98.50 ours SRC+SRLP(I,PI,HOGs) 98.31 sermanet [27] Multi-scale CNNs 98.19 [28] SRGE+HOG2 96.14 CAOR [4] Random ... |

12 |
Gool, Sparse representation based projections
- Timofte, Van
(Show Context)
Citation Context ...gn Detection Benchmark (GTSD) [9]1, the German Traffic Sign Recognition Benchmark (GTSC) [4] and the Belgium Traffic Sign Dataset with its split for Detection (BTSD) [3] and for Classification (BTSC) =-=[10]-=-2. The choice of the datasets is motivated by their large amount of annotations, diversity of the content and classes, the availability of a split for benchmarking separately TSD and TSC. The GTSD rec... |

11 | Detection of Traffic Signs in Real-World Images: The German Traffic Sign Detection Benchmark.
- Houben, Stallkamp, et al.
- 2013
(Show Context)
Citation Context ...false positives from either dataset. II. TRAFFIC SIGN DATASETS In this paper we evaluate traffic sign detection and classification on four datasets: the German Traffic Sign Detection Benchmark (GTSD) =-=[9]-=-1, the German Traffic Sign Recognition Benchmark (GTSC) [4] and the Belgium Traffic Sign Dataset with its split for Detection (BTSD) [3] and for Classification (BTSC) [10]2. The choice of the datasets... |

10 |
vs. computer: Benchmarking machine learning algorithms for traffic sign recognition
- Stallkamp, Schlipsing, et al.
- 2012
(Show Context)
Citation Context ...r settings as in [15], where top results were achieved for handwritten digit and face classification. PI provides a 2172dimensional descriptor. HOG1, HOG2, HOG3 : HOG features as precomputed for GTSC =-=[4]-=-. Three settings are provided, the difference among them is given by the number of HOG cells and the extraction grid. HOG1 and HOG2 are 1568-dimensional, while HOG3 is 2916-dimensional. B. Dimensional... |

9 | Gool. Iterative nearest neighbors for classification and dimensionality reduction
- Timofte, Van
(Show Context)
Citation Context ...en in [5]. A. Contributions The contribution of this work is three-fold: • We show that top performance can be reached using existing approaches for pedestrian detection [6] and face recognition [7], =-=[8]-=-, without the need to encode traffic sign specific information. • We provide an extensive evaluation on two large Belgian and German traffic sign benchmarks for both detection and classification. • We... |

7 |
Vision based Traffic Sign Detection and Analysis for Intelligent Driver Assistance Systems
- Møgelmose, Trivedi, et al.
- 2012
(Show Context)
Citation Context ...beling of such detections into specific traffic sign types or subcategories. For TSD and TSC numerous approaches have been developed. A recent survey of such methods and existing datasets is given in =-=[5]-=-. A. Contributions The contribution of this work is three-fold: • We show that top performance can be reached using existing approaches for pedestrian detection [6] and face recognition [7], [8], with... |

3 |
Fast and accurate digit classification,” EECS
- Maji, Malik
- 2009
(Show Context)
Citation Context ...opped traffic sign images rescaled to 28 × 28 pixels. The I features are 784dimensional. PI: the pyramid of histograms of oriented gradients (HOG) features with their optimal parameter settings as in =-=[15]-=-, where top results were achieved for handwritten digit and face classification. PI provides a 2172dimensional descriptor. HOG1, HOG2, HOG3 : HOG features as precomputed for GTSC [4]. Three settings a... |

2 |
Sparse-representation-based graph embedding for traffic sign recognition
- Lu, Ding, et al.
(Show Context)
Citation Context ...C CR (%) Team Method 99.46 IDSIA [26] Committee of CNNs 98.84 INI-RTCV [4] Human Performance 98.53 ours INNC+INNLP(I,PI,HOGs) 98.50 ours SRC+SRLP(I,PI,HOGs) 98.31 sermanet [27] Multi-scale CNNs 98.19 =-=[28]-=- SRGE+HOG2 96.14 CAOR [4] Random Forests 95.68 INI-RTCV [4] LDA on HOG2 The same effect is found on BTSC. Concatenating LDA projections of I and of PI features boosts the performance up to 97.78% for ... |

1 |
Chapter 3.5: Combining traffic sign detection with 3d tracking towards better driver assistance
- Timofte, Prisacariu, et al.
- 2011
(Show Context)
Citation Context ... SIGN RECOGNITION (TSR) gets consider-able interest lately. The interest is driven by the market for intelligent applications such as autonomous driving [1], advanced driver assistance systems (ADAS) =-=[2]-=-, mobile mapping [3], and the recent releases of larger traffic signs datasets such as Belgian [3] or German [4] datasets. TSR covers two problems: traffic sign detection (TSD) and traffic sign classi... |