Projects

Multimodal learning (Multimodality data + Deep learning)

Cross-domain Deep Feature Combination for Bird Species Classification with Audio-visual Data

  • Purpose: to classify bird species
  • Background: difficulty of fine-grained classification + employ art/pose based approaches to achieve good performance
  • Methodology: Audio-Visual data + CNN + Multimodal learning

Product Detection on Retail Store Shelves

Dividing Products in Store Shelves by Detecting Horizontal Clapboard and Shadow for Product Recognition

Segmentation based Non-learning Product Detection for Product Recognition on Store Shelves

 

  • Purpose: to detect products on retail store shelves
  • Background: manual managements of the product arrangement waste resource + present methods require a template of the product
  • Methodology: use segmentation to detect products without the templates

Multi-view Face Detection

Multi-view Face Detection using Flipping Scheme

Multi-view Face Detection using Frontal Face Detector

  • Purpose: multi-view face detection
  • Background: complex logic structure for multi-view detector
  • Methodology: use frontal face detector only + mirror reversal theory

Rat head gaze determination

A Novel Rat Head Gaze Determination System Based on Optomotor Responses

  • Purpose: rat head gaze determination
  • Background: difficult to evaluate rat behavior without markers
  • Methodology: homogeneous filtering + head gaze extraction
  • Takuya Akashi, Hitoshi Kubota, Hiroyuki Tomita, Hiroyuki Konno.

Face sensing

High-speed Multiview Face Localization and Tracking with a Minimum Bounding Box Using Genetic Algorithm

  • Purpose: multi-view face detection & tracking
  • Background: high effort & cost of creating dataset & machine learning
  • Methodology: no prior machine learning + evolutionary video processing by GA

Human pose estimation

3D Skeleton Matters: Human Pose Retrieval by Human Pose Reconstruction

  • Haitian Sun, Chao Zhang, Takuya Akashi.

Interactive Depth Ambiguity Correction for 3D Human Pose Estimation

  • Hiroki Hirayama, Haitian Sun, Chao Zhang, Takuya Akashi.

  • Junya Sato, Takuya Akashi.

Parametric template matching

Fast Affine Template Matching over Galois Field

  • Chao Zhang, Takuya Akashi.

Robust Projective Template Matching

Full results comparing against SIFT and ASIFT

  • Chao Zhang, Takuya Akashi.
  • Junya Sato, Takuya Akashi.

3D Pose Estimation

Cube Detection and Pose Estimation Using Property of a Cube with Evolutionary Method

  • Kazuki Hiyama, Takuya Akashi.

Online Visual Tracking

Robust Visual Tracking via Coupled Randomness

  • Chao Zhang, Takuya Akashi.

Local-changes invariant image matching

High-speed and Local-changes Invariant Image Matching

  • Chao Zhang, Takuya Akashi.

Sketch-based image retrieval

Sketch-based Image Retrieval:Weighted HOG By Grayscale Value

  • Sun Haitian, Zhang Chao, Takuya Akashi.

  • Shigenobu Kato, Akio Numakura, Kazuyuki Sato, Tasuku Miyoshi, Kim Chyon Hae, Takuya Akashi.

Traffic sensing

Sensing of traffic flow in real-world for reconstruction in VR

  • Guo Tian, Takuya Akashi.

  • 佐藤遼, 日山一樹, 明石卓也.

Distortion correction

キャリブレーション不要な進化的歪み補正に関する基礎研究

  • 引屋敷和博, 佐藤惇哉, 明石卓也.

Basic shape detection

Deterministic Crowdingによる円検出に関する研究

  • 平野靖尚, 佐藤惇哉, 明石卓也.

Road safety

Detection of Rushing out into the Street Using the Change of Scene

  • Shinnosuke Naito, Takuya Akashi.

Pedestrian sensing with first-person view

An Evolutional Pedestrian Detection Method with Probability-Based Model

  • Chao Zhang, Takuya Akashi.

Other Applied Research

Beauty Expert System using Face Sensing

  • A-STEP High-risk Challenging type (Revitalization Promotion type) Type I, JST, 2014-2015.

Other Applied Research

  • Revitalization Promotion Program (Matching Promotion) Type II, JST, 2013-2015.