Yuji Yamauchi

People detection based on co-occurrence of appearance and spatiotemporal features

This paper presents a method for detecting people based on the co-occurrence of appearance and spatiotemporal features. Histograms of oriented gradients(HOG) are used as appearance features, and the results of pixel state analysis are used as spatiotemporal features. The pixel state analysis classifies foreground pixels as either stationary or transient. The appearance and spatiotemporal features are projected into subspaces in order to reduce the dimensions of the vectors by principal component analysis(PCA). The cascade AdaBoost classifier is used to represent the co-occurrence of the appearance and spatiotemporal features. The use of feature co-occurrence, which captures the similarity of appearance, motion, and spatial information within the people class, makes it an effective detector. Experimental results show that the performance of our method is about 29% better than that of the conventional method.

Pedestrian detection based on deep convolutional neural network with ensemble inference network

Pedestrian detection is an active research topic for driving assistance systems. To install pedestrian detection in a regular vehicle, however, there is a need to reduce its cost and ensure high accuracy. Although many approaches have been developed, vision-based methods of pedestrian detection are best suited to these requirements. In this paper, we propose the methods based on Convolutional Neural Networks (CNN) that achieves high accuracy in various fields. To achieve such generalization, our CNN-based method introduces Random Dropout and Ensemble Inference Network (EIN) to the training and classification processes, respectively. Random Dropout selects units that have a flexible rate, instead of the fixed rate in conventional Dropout. EIN constructs multiple networks that have different structures in fully connected layers. The proposed methods achieves comparable performance to state-of-the-art methods, even though the structure of the proposed methods are considerably simpler.

Relational HOG feature with wild-card for object detection

This paper proposes Relational HOG (R-HOG) features for object detection, and binary selection by using a wild-card “*” with Real AdaBoost. HOG features are effective for object detection, but their focus on local regions makes them high-dimensional features. To reduce the memory required for the HOG features, this paper proposes a new feature, R-HOG, which creates binary patterns from the HOG features extracted from two local regions. This approach enables the created binary patterns to reflect the relationships between local regions. Furthermore, we extend the R-HOG features by shifting the gradient orientations. These shifted Relational HOG (SR-HOG) features make it possible to clarify the size relationships of the HOG features. However, since R-HOG and SR-HOG features contain binary values not needed for classification, we have added a process to the Real AdaBoost learning algorithm in which “*” permits either of the two binary values “0” and “1”, and so valid binary values can be selected. Evaluation experiment demonstrated that the SR-HOG features introducing “*” offers better detection performance than the conventional method (HOG feature) despite the reduced memory requirements.

To Be Bernoulli or to Be Gaussian, for a Restricted Boltzmann Machine

We introduce a method that automatically selects appropriate RBM types according to the visible unit distribution. The distribution of a visible unit strongly depends on a dataset. For example, binary data can be considered as pseudo binary distribution with high peaks at 0 and 1. For real-value data, the distribution can be modeled by single Gaussian model or Gaussian mixture model. Our proposed method selects appropriate RBM according to the distribution of each unit. We employ the Gaussian mixture model to determine whether the visible unit distribution is the pseudo binary or the Gaussian mixture. According to this distribution, we can select a Bernoulli-Bernoulli RBM(BBRBM) or a Gaussian-Bernoulli RBM(GBRBM). Furthermore, we employ normalization process to obtain a smoothed Gaussian mixture distribution. This allowed us to reduce variations such as illumination changes in the input data. After experimentation with MNIST, CBCL and our own dataset, our proposed method obtained the best recognition performance and further shortened the convergence time of the learning process.

CS-HOG: Color similarity-based HOG

Conventional object detection methods often use local features based on object shape, of which the HOG feature is typical. In recent years, Color Self-Similarity (CSS) has been proposed as a local feature that uses color information. CSS involves computing color similarity as a basis for deciding the sameness of objects, and thus represent a feature that is effective for object detection. It has also been reported that detection performance can be improved by using the CSS feature together with the HOG feature or other shape-based feature. We propose a Color Similarity-based HOG (CS-HOG) feature that is based on color similarity for object detecting shapes. The CS-HOG feature enables clarification of the object shape by using color similarity to calculate the degree of object sameness, thus achieving highly-accurate object detection. Evaluation experiments show that the CS-HOG feature improves performance from 22.5% and 27.2% compared to the HOG feature and the CSS feature, and by 4.2% compared to the HOG feature and the CSS feature used together.

Feature co-occurrence representation based on boosting for object detection

This paper proposes a method of feature co-occurrence representation based on boosting for object detection. A previously proposed method that combines multiple binary-classified codes by AdaBoost to represent the co-occurrence of features has been shown to be effective in face detection. However, if an input feature is difficult to be assigned to a correct binary code due to occlusion or other factors, a problem arises here since the process of binary classification and co-occurrence representation may combine features that include an erroneous code. In response to this problem, this paper proposes a Co-occurrence Probability Feature (CPF) that combines multiple weak classifiers by addition and multiplication arithmetic operators using Real AdaBoost in which the outputs of weak classifiers are real values. Since CPF combines classifiers using two types of operators, diverse types of co-occurrence can be represented and improved detection performance can be expected. To represent even more diversified co-occurrence, this paper also proposes co-occurrence representation that applies a subtraction arithmetic operator. Although co-occurrence representation using addition and multiplication operators can represent co-occurrence between features, use of the subtraction operator enables the representation of co-occurrence between local features and features having other properties. This should have the effect of revising the probability of the detection-target class obtained from local features. Evaluation experiments have shown co-occurrence representation by the proposed methods to be effective.

Cost-alleviative Learning for Deep Convolutional Neural Network-based Facial Part Labeling

Facial part labeling which is parsing semantic components enables high-level facial image analysis, and contributes greatly to face recognition, expression recognition, animation, and synthesis. In this paper, we propose a cost-alleviative learning method that uses a weighted cost function to improve the performance of certain classes during facial part labeling. As the conventional cost function handles the error in all classes equally, the error in a class with a slightly biased prior probability tends not to be propagated. The weighted cost function enables the training coefficient for each class to be adjusted. In addition, the boundaries of each class may be recognized after fewer iterations, which will improve the performance. In facial part labeling, the recognition performance of the eye class can be significantly improved using cost-alleviative learning.

Keypoint detection by cascaded fast

When the FAST method for detecting corner features at high speed is applied to images that include complex textures (regions that include foliage, shrubbery, etc.), many corners that are not needed for object recognition are detected because FAST defines corner features on the basis of a 16-pixel bounding circle. To overcome that problem, we propose the Cascaded FAST that defines corners on the basis of similarity in terms of intensity, continuity and orientation in a broader range of areas (20, 16, and 12 pixel bounding circles). Also, cascading three decision trees trained by the FAST approach enables high-speed corner detection in which non-corners are eliminated early in the process. Furthermore, Cascaded FAST determines scale by using an image pyramid and determines orientation at high speed by using a framework for referencing surrounding pixels.

Facial point detection based on a convolutional neural network with optimal mini-batch procedure

We propose a Convolutional Neural Network (CNN)-based method to ensure both robustness to variations in facial pose and real-time processing. Although the robustness of CNNs has attracted attention in various fields, the training process suffers from difficulties in parameter setting and the manner in which training samples are provided. We demonstrate a manner of providing samples that results in a better network. We consider four methods: 1) subset with augmentation, 2) random selection, 3) fixed-person subset, and 4) the conventional approach. Experimental results indicate that the subset with augmentation technique has sufficient variations and quantity to obtain the best performance. Our CNN-based method is robust under facial pose variations, and achieves better performance. In addition, since our networks structure is simple, processing takes approximately 10ms for one face on a standard CPU.

Fast 3D edge detection by using decision tree from depth image

T3D edge detection from a depth image is an important technique of 3D object recognition in preprocessing. There are three types of 3D edges in a depth image called jump, convex roof, and concave roof edges. Conventional 3D edge detection based on ring operators has been proposed. The conventional ring operator can detect three types of 3D edges by classifying the response of Fourier transforms. Since the conventional method needs to apply Fourier transforms to all pixels of a depth image, real-time processing cannot be done due to high computational cost. Therefore, this paper presents a fast and reliable method of detecting three types of 3D edges by using a decision tree. The decision tree is trained under supervised learning from numerous synthesized depth images and labels by capturing depth relations between candidate pixels and pixels on a ring operator to classify 3D edges. The experimental results revealed that the proposed method has 25 times faster than the conventional method. This paper also presents some examples of 3D line and 3D convex corner detection based on results obtained with the proposed method.