Shuichi Akizuki

DPN-LRF: A local reference frame for robustly handling density differences and partial occlusions

For the purpose of 3D keypoint matching, a Local Reference Frame (LRF), a local coordinate system of the keypoint, is one important information source for achieving repeatable feature descriptions and accurate pose estimations. We propose a robust LRF for two main point cloud disturbances: density differences and partial occlusions. To generate LRFs that are robust to such disturbances, we employ two strategies: normalizing the effects of point cloud density by approximating the surface area in the local region and using the dominant orientation of a normal vector around the keypoint. Experiments confirm that the proposed method has higher repeatability than state-of-the-art methods with respect to density differences and partial occlusions. It was also confirmed that the method enhances the reliability of keypoint matching.

SHORT: A fast 3D feature description based on estimating occupancy in spherical shell regions

We propose a novel feature description method called SHORT (Shell Histograms and Occupancy from Radial Transform) for fast 3D object recognition. In 3D object recognition for point cloud data, it is very important to detect keypoints and describe features rapidly because of the huge amount of data involved. The state-of-the-art keypoint detection methods calculate statistics including covariance matrices from the point cloud in local regions of the object. Then, the state-of-the-art method, which describe features such as normal vector distributions of the point cloud, use all points in the local regions. However, these methods involve high processing costs because they need to calculate the statistics needed for keypoint detection. They also need to use a lot of points in the regions for feature description. By contrast, the SHORT method consists of a fast keypoint detector that does not calculate statistics and a fast feature descriptor that uses only a small number of points in the restricted local regions. The keypoint detector uses occupancy estimated simply like counting the points in regions of outermost shells in spheres, and the feature descriptor uses estimated those and a small number of points including the spherical shell regions of multiple scales. Experimental results in 3D object recognition show that the processing speed of the proposed method is five times faster than that of a comparative method that had a nearly equal 99.4% recognition success rate.

High-speed and reliable object recognition using distinctive 3-D vector-pairs in a range image

We propose a high-speed 3-D object detection method that can recognize the position and pose of objects in complicated scenes consisting of randomly stacked objects. The methods main feature is that a set of distinctive 3-D vector pairs, each of which consists of three different 3-D points, is used for matching objects with an acquired range image. Such distinctive vector pairs represent the local shape of an object, and are extracted by calculating the occurrence probability of each 3-D vector pair in a model object. A vector pair with a low occurrence probability means that it is distinctive not only in a model but also in an acquired image. Therefore, the method is expected to avoid false matching even if there are similar objects around the target object. It also substantially reduces processing time because the number of vector pairs is much smaller than all the data points of the object model. Experiments confirm that in comparison with the Spin Image method, the proposed method is about 60 times faster and increases the recognition success rate from 62.0% to 94.6%.

3-D Feature Point Matching for Object Recognition Based on Estimation of Local Shape Distinctiveness

In this paper, we propose a reliable 3-D object recognition method that can statistically minimize object mismatching. Our method basically uses a 3-D object model that is represented as a set of feature points with 3-D coordinates. Each feature point also has an attribute value for the local shape around the point. The attribute value is represented as an orientation histogram of a normal vector calculated by using several neighboring feature points around each point. Here, the important thing is this attribute value means its local shape. By estimating the relative similarity of two points of all possible combinations in the model, we define the distinctiveness of each point. In the proposed method, only a small number of distinctive feature points are selected and used for matching with all feature points extracted from an acquired range image. Finally, the position and pose of the target object can be estimated from a number of correctly matched points. Experimental results using actual scenes have demonstrated that the recognition rate of our method is 93.8%, which is 42.2% higher than that of the conventional Spin Image method. Furthermore, its computing time is about nine times faster than that of the Spin Image method.

Affordance-based 3D feature for generic object recognition

Techniques for generic object recognition, which targets everyday objects such as cups and spoons, and techniques for approach vector estimation (e.g. estimating grasp position), which are needed for carrying out tasks involving everyday objects, are considered necessary for the perceptual system of service robots. In this research, we design feature for generic object recognition so they can also be applied to approach vector estimation. To carry out tasks involving everyday objects, estimating the function of the target object is critical. Also, as the function of holding liquid is found in all cups, so a function is shared in each type (class) of everyday objects. We thus propose a generic object recognition method that can estimate the approach vector by expressing an object’s function as feature. In a test of the generic object recognition of everyday objects, we confirmed that our proposed method had a 92% recognition rate. This rate was 11% higher than the mainstream generic object recognition technique of using convolutional neural network (CNN).

Multiple 3D Object Recognition using RGB-D Data and Physical Consistency for Automated Warehousing Robots

In this research, we propose a method to recognize multiple objects in the shelves of automated warehouses. The purpose of this research is to enhance the reliability of the Hypothesis Verification (HV) method that simultaneously recognizes layout of multiple objects. The proposed method have employed not only the RGB-D consistency between the input scene and the scene hypothesis but also the physical consistency. By considering the physical consistency of the scene hypothesis, the proposed HV method can efficiently reject false one. Experiment results for object which are used at Amazon Picking Challenge 2015 have been confirmed that the recognition success rate of the proposed method is higher than the previous HV method.

Physical Reasoning for 3D Object Recognition Using Global Hypothesis Verification

In this paper, we propose a method to recognize the 6DoF pose of multiple objects simultaneously. One good solution to recognize them is applying a Hypothesis Verification (HV) algorithm. This type of algorithm evaluates consistency between an input scene and scene hypotheses represented by combinations of object candidates generated from the model based matching. Its use achieves reliable recognition because it maximizes the fitting score between the input scene and the scene hypotheses instead of maximizing the fitting score of an object candidate. We have developed a more reliable HV algorithm that uses a novel cue, the naturalness of an object’s layout (its physical reasoning). This cue evaluates whether the object’s layout in a scene hypothesis can actually be achieved by using simple collision detection. Experimental results show that using the physical reasoning have improved recognition reliability.

Reliable image matching using binarized gradient features obtained with multi-flash camera

In this paper, we propose an object detection method using features describing information about a concavoconvex shape of an object that are obtained by using a small camera that controls the illumination direction. A feature image containing information about the shape of the object is generated by integrating images obtained by turning on, one by one, light emitting diodes (LEDs) annularly arranged around the camera. Our method can reliably detect a texture-less object by using this feature image in the matching process. Experiments using 200 actual images confirmed that the method achieves a 97.5% recognition success rate and a 4.62 sec processing time.

High-speed and reliable object recognition based on low-dimensional local shape features

In this paper, we propose a high-speed 3-D object recognition method using new feature values. Features for the object recognition method proposed in this study consist of three values. One is the Difference of Normals (DoN) feature value that has been proposed by Ioannou. The other two represent information about curvature. We use these three-dimensional features to recognize the position and pose of multiple objects stacked randomly. Because they are low-dimensional, high-speed matching can be achieved. We have also reduced the computing time needed for data matching by using only effective points selected on the basis of their estimated distinctiveness. Experimental results using actual scenes have demonstrated that the computing time is about 93 times faster than that of the conventional SHOT method. Furthermore, the proposed method achieves a 98.2% recognition rate, which is 17.9% higher than that of the SHOT method. Also, we confirmed that the proposed method achieves higher-speed matching and higher recognition success rate than the conventional methods.