Local Feature Based Multiple Object Instance Identification Using Scale and Rotation Invariant Implicit Shape Model

Bao, Ruihan; Higa, Kyota; Iwamoto, Kota

doi:10.1007/978-3-319-16628-5_43

Cited by 6 publications

(4 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Satapathy et al [13] proposed a method to detect products on the shelves by using exhaustive template matching for raising alerts to replenish products. Bao et al [14] proposed a method to accurately identify and locate a large number of objects in an image by using a scalable and compact binary local descriptor, named the BRIGHT (Binary ResIzable Gradient HisTogram) descriptor [15]. Zhang et al [16] proposed a dual-layer density estimation-based architecture to detect multiple object instances from an image for robot inventory management.…”

Section: Related Workmentioning

confidence: 99%

“…Scale and rotation invariant local descriptors, such as SIFT, SURF and BRIGHT, have enabled identification of real-world objects in images with robustness against differing camera viewpoints, occlusions and lighting conditions. To achieve high performance, these methods [3,12,13,14,16,18] require a large database of high-quality product templates that must be updated every time products on the shelves change. However, the large database is time-consuming to maintain since the products on the shelves frequently change due to new and seasonal products arriving.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Robust Shelf Monitoring Using Supervised Learning for Improving On-Shelf Availability in Retail Stores

Higa

Iwamoto

2019

Sensors

Self Cite

View full text Add to dashboard Cite

This paper proposes a method to robustly monitor shelves in retail stores using supervised learning for improving on-shelf availability. To ensure high on-shelf availability, which is a key factor for improving profits in retail stores, we focus on understanding changes in products regarding increases/decreases in product amounts on the shelves. Our method first detects changed regions of products in an image by using background subtraction followed by moving object removal. It then classifies the detected change regions into several classes representing the actual changes on the shelves, such as “product taken (decrease)” and “product replenished/returned (increase)”, by supervised learning using convolutional neural networks. It finally updates the shelf condition representing the presence/absence of products using classification results and computes the product amount visible in the image as on-shelf availability using the updated shelf condition. Three experiments were conducted using two videos captured from a surveillance camera on the ceiling in a real store. Results of the first and second experiments show the effectiveness of the product change classification in our method. Results of the third experiment show that our method achieves a success rate of 89.6% for on-shelf availability when an error margin is within one product. With high accuracy, store clerks can maintain high on-shelf availability, enabling retail stores to increase profits.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Robust Shelf Monitoring Using Supervised Learning for Improving On-Shelf Availability in Retail Stores

Higa

Iwamoto

2019

Sensors

Self Cite

View full text Add to dashboard Cite

show abstract

“…In the second place, we estimate the object center o ik ' x oik ' , y oik ' for every valid key point p ik ' x ik ' , y ik ' in the query image for the matching training image i, where i . Reference centers corresponding to the matched training image can be calculated on the basis of formula (2) and (3) [10][11][12]. In the formulas, s ik ' and ik ' are the corresponding scale and orientation of the key point p ik ' .…”

Section: False Matches Elimination Stagementioning

confidence: 99%

An improved method for object instance detection based on object center estimation and convex quadrilateral verification

Zhang

et al. 2016

2016 IEEE Information Technology, Networking, Electronic and Automation Control Conference

View full text Add to dashboard Cite

This paper proposes a new frame which combines kernel density estimation with convex quadrilateral verification to achieve identification and localization of object instance in zooming image. Reference object centers are first calculated based on scales, orientations, and reference vectors for all matched key points. Then the valid object center with density peak value is held. A region of interest with evaluated radius is confirmed based on the scale of the center point and row and column number of the corresponding training image. Homography matrix is computed using the matched key points within the region. Finally, four edges of the candidate training image are mapped as a quadrilateral. Only if the quadrilateral is convex that the object instance is detected. The experimental results prove that the proposed approach provides high efficiency for real-time applications with robustness.

show abstract

“…However, the key point coordinates obtained from the clustering results in [13][14][15] might be unreliable because the key points are sparsely distributed. Alternatively, approaches based on Hough voting were proposed and applied in [16][17][18]. The Hough voting based approach locates possible instances according to feature mapping and density estimation.…”

Section: Introductionmentioning

confidence: 99%

Dual-Layer Density Estimation for Multiple Object Instance Detection

Zhang

et al. 2016

Journal of Sensors

View full text Add to dashboard Cite

This paper introduces a dual-layer density estimation-based architecture for multiple object instance detection in robot inventory management applications. The approach consists of raw scale-invariant feature transform (SIFT) feature matching and key point projection. The dominant scale ratio and a reference clustering threshold are estimated using the first layer of the density estimation. A cascade of filters is applied after feature template reconstruction and refined feature matching to eliminate false matches. Before the second layer of density estimation, the adaptive threshold is finalized by multiplying an empirical coefficient for the reference value. The coefficient is identified experimentally. Adaptive threshold-based grid voting is applied to find all candidate object instances. Error detection is eliminated using final geometric verification in accordance with Random Sample Consensus (RANSAC). The detection results of the proposed approach are evaluated on a self-built dataset collected in a supermarket. The results demonstrate that the approach provides high robustness and low latency for inventory management application.

show abstract

Local Feature Based Multiple Object Instance Identification Using Scale and Rotation Invariant Implicit Shape Model

Cited by 6 publications

References 17 publications

Robust Shelf Monitoring Using Supervised Learning for Improving On-Shelf Availability in Retail Stores

Robust Shelf Monitoring Using Supervised Learning for Improving On-Shelf Availability in Retail Stores

An improved method for object instance detection based on object center estimation and convex quadrilateral verification

Dual-Layer Density Estimation for Multiple Object Instance Detection

Contact Info

Product

Resources

About