A Novel Method for Approximating Object Location Error in Bounding Box Detection Algorithms Using a Monocular Camera

Abstract. In this research, a structured illumination reflectance imaging (SIRI) system was used as a novel method for detection of fresh bruises on apples. The SIRI system projects sinusoidal patterns of illumination onto samples, and image demodulation is then used to recover depth-specific information through varying the spatial frequency of the illumination pattern. The capability of SIRI was demonstrated through the detection of artificially induced bruises on ‘Golden Delicious’ and ‘Delicious’ apples with varying levels of bruising. It was hypothesized that by confining the light penetration depth near the surface of each fruit, subsurface defects such as bruising should be more apparent under SIRI than conventional planar illumination imaging. Three 120° phase-shifted reflectance images were acquired from 60 fruit each of ‘Golden Delicious’ and ‘Delicious’ varieties at 0 h, 4 to 6 h, and 24 h after impact bruising for each of the four spatial frequencies (i.e., 0, 0.10, 0.15, and 0.25 cycles mm-1). The reflectance images acquired by the system were then demodulated into an alternating component (AC) and direct component (DC), where the AC contained depth-specific information and the DC image represented the diffuse reflectance from the apple sample under uniform (or planar) illumination. Bruise detection algorithms were developed and applied to the demodulated AC and DC images. The SIRI system achieved 70% to 100% bruise detection rates, compared to 0% to 50% detection rates under conventional planar illumination. However, detection results were influenced by both severity of bruising and bruise development after impact; better bruise detection results were obtained when bruises had developed for 4 to 6 h after impact. SIRI has demonstrated a superior capability of detecting fresh bruises, and it is promising as a new imaging modality for quality detection of agricultural products. Keywords: Apples, Bruises, Defects, Fruit, Imaging, Nondestructive, Quality, Sorting, Structured illumination.

Download Full-text

All-sky photogrammetry techniques to georeference a cloud field

Atmospheric Measurement Techniques ◽

10.5194/amt-11-593-2018 ◽

2018 ◽

Vol 11 (1) ◽

pp. 593-609 ◽

Cited By ~ 5

Author(s):

Pierre Crispel ◽

Gregory Roberts

Keyword(s):

A Priori ◽

Velocity Fields ◽

Contour Detection ◽

Cloud Base ◽

Reconstruction Technique ◽

Cumulus Clouds ◽

Detection Algorithms ◽

Camera Orientation ◽

Novel Method ◽

Cloud Base Height

Abstract. In this study, we present a novel method of identifying and geolocalizing cloud field elements from a portable all-sky camera stereo network based on the ground and oriented towards zenith. The methodology is mainly based on stereophotogrammetry which is a 3-D reconstruction technique based on triangulation from corresponding stereo pixels in rectified images. In cases where clouds are horizontally separated, identifying individual positions is performed with segmentation techniques based on hue filtering and contour detection algorithms. Macroscopic cloud field characteristics such as cloud layer base heights and velocity fields are also deduced. In addition, the methodology is fitted to the context of measurement campaigns which impose simplicity of implementation, auto-calibration, and portability. Camera internal geometry models are achieved a priori in the laboratory and validated to ensure a certain accuracy in the peripheral parts of the all-sky image. Then, stereophotogrammetry with dense 3-D reconstruction is applied with cameras spaced 150 m apart for two validation cases. The first validation case is carried out with cumulus clouds having a cloud base height at 1500 m a.g.l. The second validation case is carried out with two cloud layers: a cumulus fractus layer with a base height at 1000 m a.g.l. and an altocumulus stratiformis layer with a base height of 2300 m a.g.l. Velocity fields at cloud base are computed by tracking image rectangular patterns through successive shots. The height uncertainty is estimated by comparison with a Vaisala CL31 ceilometer located on the site. The uncertainty on the horizontal coordinates and on the velocity field are theoretically quantified by using the experimental uncertainties of the cloud base height and camera orientation. In the first cumulus case, segmentation of the image is performed to identify individuals clouds in the cloud field and determine the horizontal positions of the cloud centers.

Download Full-text

Bolt damage identification based on orientation-aware center point estimation network

Structural Health Monitoring ◽

10.1177/14759217211004243 ◽

2021 ◽

pp. 147592172110042

Author(s):

Yang Zhang ◽

Ka-Veng Yuen

Keyword(s):

Neural Network ◽

Damage Identification ◽

Deep Layer ◽

Point Estimation ◽

Learning Object ◽

Detection Algorithms ◽

Network Orientation ◽

Center Point ◽

Bounding Box ◽

Optimization Efficiency

With the development of deep learning, object detection algorithms based on horizontal box are widely used in the field of damage identification. However, damages can be in any direction and position, and they are not necessarily horizontal or vertical. This article proposes a bolt damage identification network, namely, orientation-aware center point estimation network, which models a damage as a center point of its rotated bounding box. The proposed orientation-aware center point estimation network uses deep layer aggregation network to search center points and regress to all other damage properties, such as size and angle. A loss function is designed to improve the optimization efficiency of network. Orientation-aware center point estimation network is applied to bolt damage detection, and comparison with the well-known Faster Region-Convolutional Neural Network (a benchmark using horizontal bounding box) demonstrates the accuracy of the proposed method. Finally, videos were utilized to verify the capability of the proposed orientation-aware center point estimation network in real-time detection of bolt damages.

Download Full-text

Distance-IoU Loss: Faster and Better Learning for Bounding Box Regression

Proceedings of the AAAI Conference on Artificial Intelligence ◽

10.1609/aaai.v34i07.6999 ◽

2020 ◽

Vol 34 (07) ◽

pp. 12993-13000 ◽

Cited By ~ 15

Author(s):

Zhaohui Zheng ◽

Ping Wang ◽

Wei Liu ◽

Jinze Li ◽

Rongguang Ye ◽

...

Keyword(s):

Object Detection ◽

Performance Improvement ◽

Detection Algorithms ◽

Art Object ◽

Bounding Box ◽

Geometric Factors ◽

Evaluation Metric ◽

Point Distance ◽

Performance Gains ◽

Overlap Area

Bounding box regression is the crucial step in object detection. In existing methods, while ℓn-norm loss is widely adopted for bounding box regression, it is not tailored to the evaluation metric, i.e., Intersection over Union (IoU). Recently, IoU loss and generalized IoU (GIoU) loss have been proposed to benefit the IoU metric, but still suffer from the problems of slow convergence and inaccurate regression. In this paper, we propose a Distance-IoU (DIoU) loss by incorporating the normalized distance between the predicted box and the target box, which converges much faster in training than IoU and GIoU losses. Furthermore, this paper summarizes three geometric factors in bounding box regression, i.e., overlap area, central point distance and aspect ratio, based on which a Complete IoU (CIoU) loss is proposed, thereby leading to faster convergence and better performance. By incorporating DIoU and CIoU losses into state-of-the-art object detection algorithms, e.g., YOLO v3, SSD and Faster R-CNN, we achieve notable performance gains in terms of not only IoU metric but also GIoU metric. Moreover, DIoU can be easily adopted into non-maximum suppression (NMS) to act as the criterion, further boosting performance improvement. The source code and trained models are available at https://github.com/Zzh-tju/DIoU.

Download Full-text

An Insight into Traffic Light Discernment and Cognizance using Support Vector Machine, Multi Class Learning and Deep Learning Concepts

International Journal of Recent Technology and Engineering - 2 ◽

10.35940/ijrte.a3074.059120 ◽

2020 ◽

Vol 9 (1) ◽

pp. 2526-2534

Keyword(s):

Deep Learning ◽

Object Detection ◽

Support Vector ◽

Traffic Light ◽

Full Data ◽

Traffic Lights ◽

Detection Algorithms ◽

Bounding Box ◽

Area Unit ◽

Multiclass Classifier

This paper principally combines ideas of laptop vision, machine learning and deep learning for correct detection of traffic lights and their classifications. It checks for each circular and arrow stoplight cases. Color filtering and blob discover ion area unit principally to detect the candidates (traffic lights) [6]. Then, a PCA network is employed as a multiclass classifier which provides the result sporadically. MOT will used for more trailing method and prediction filters out false positives. Sometimes, vote theme can even be used rather than MOT. This method will be simply fitted into ADAS vehicles once hardware thinks about. Recognition is as vital as detective work the traffic lights. While not recognition, no full data will be transmitted [2]. Many complicated TLR’s will give advance functions like observing the most the most for a specific route (when there's quite one) and the way removed from the driving force [3]. Deep learning is additionally one among the rising techniques for analysis areas [7]. Object detection comes as associate integral a part of laptop vision. Object detection will be best utilized in create estimation, vehicle detection, police work etc. In detection algorithms, we tend to incline to draw a bounding box round the object of interest to find it among the image. Also, the drawing of the bounding box isn't distinctive and might hyperbolically looking on the need [9].

Download Full-text

Detection of double defects for plate-like structures based on a fuzzy c-means clustering algorithm

Structural Health Monitoring ◽

10.1177/1475921718772042 ◽

2018 ◽

Vol 18 (3) ◽

pp. 757-766 ◽

Cited By ~ 4

Author(s):

Shaojie Chen ◽

Shaoping Zhou ◽

Chaofeng Chen ◽

Yong Li ◽

Shuangmiao Zhai

Keyword(s):

Clustering Algorithm ◽

Signal To Noise Ratio ◽

Circular Array ◽

Location Error ◽

Direct Wave ◽

Fuzzy C Means ◽

Single Defect ◽

Fuzzy C Means Clustering ◽

Signal Processing Algorithms ◽

Novel Method

A variety of signal processing algorithms have been proposed to detect and locate defects in plate-like structures. However, the signal-to-noise ratio in these algorithms is too small especially in the reflection wave from the boundary, which further degrades the accuracy of localization of defects. A novel method for localization of defects is proposed in this article, based on the direct wave and fuzzy c-means clustering algorithm. To verify its effectiveness, experiments using the parallel linear and circular array are conducted, respectively. The experimental results show that the proposed method not only accurately locates single defect but also locates double defects in plate-like structures, and by comparing with the current discrete elliptic imaging algorithm, its location error of single defect is reduced from 20–25 mm to 0–3 mm and double defects is also reduced from 60–90 mm to 0–3 mm.

Download Full-text

A Novel Method for Laser Peak Detection with Subpixel Accuracy for the Rail Corrugation Measurement

Journal of Sensors ◽

10.1155/2021/6695674 ◽

2021 ◽

Vol 2021 ◽

pp. 1-16

Author(s):

Mehran Torabi ◽

S. Mohammad Mousavi G ◽

Davood Younesian

Keyword(s):

Detection Algorithm ◽

Peak Detection ◽

Prototype System ◽

Noise Elimination ◽

Rail Corrugation ◽

Detection Algorithms ◽

Laser Stripe ◽

Subpixel Accuracy ◽

Peak Detection Algorithm ◽

Novel Method

In this article, a new wavelet-based laser peak detection algorithm is proposed having subpixel accuracy. The algorithm provides an accurate and rapid measurement platform for the rail surface corrugation with no need to any image noise elimination. The proposed rail Corrugation Measurement System (CMS) is based on the laser triangulation principle, and the accuracy of such system is mainly affected by the laser peak detection in the captured image. The intensity of each row or column of the image is taken as a 1-D discrete signal. Intensity distribution of a laser stripe in this signal follows a Gaussian pattern contaminated by the white noise. Against usual peak detection algorithms with need to prenoise-filtering process, the proposed method based on the wavelet transform is able to perform these tasks efficiently and robustly. Present wavelet-based methods for the peak detection are at pixel level, but for achieving high accuracy subpixel detection is proposed. Experiments show that the capability of the proposed method for laser peak detection is more accurate and faster than the filter-based methods, especially for low S/N ratios. Also, this technique can be utilized for any application in laser peak detection with subpixel accuracy. A prototype system based on the proposed method for the rail corrugation measurement has been designed and manufactured. Results of the rail corrugation measurement guarantee capability of the proposed methodology for accurate measurement of the rail corrugation and its potential for industrial application.

Download Full-text

Machine Learning-Based Anomaly Detection Algorithms to Alert Patients Using Sensor Augmented Pump of Infusion Site Failures

Journal of Diabetes Science and Technology ◽

10.1177/1932296821997854 ◽

2021 ◽

pp. 193229682199785

Author(s):

Lorenzo Meneghetti ◽

Eyal Dassau ◽

Francis J. Doyle ◽

Simone Del Favero

Keyword(s):

Machine Learning ◽

Anomaly Detection ◽

Machine Learning Algorithms ◽

Infusion Site ◽

Detection Algorithms ◽

Quality Of Therapy ◽

Unsupervised Anomaly Detection ◽

Novel Method

Background: Personal insulin pumps have shown to be effective in improving the quality of therapy for people with type 1 diabetes (T1D). However, the safety of this technology is limited by the possible infusion site failures, which are linked with hyperglycemia and ketoacidosis. Thanks to the large availability of collected data provided by modern therapeutic technologies, machine learning algorithms have the potential to provide new way to identify failures early and avert adverse events. Methods: A clinical dataset ( N = 20) is used to evaluate a novel method for detecting real-time infusion site failures using unsupervised anomaly detection algorithms, previously proposed and developed on in-silico data. An adapted feature engineering procedure is introduced to make the method able to operate in the absence of a closed-loop (CL) system and meal announcements. Results: In the optimal configuration, we obtained a performance of 0.75 Sensitivity (15 out of 20 total failures detected) and 0.08 FP/day, outperforming previously proposed literature algorithms. The algorithm was able to anticipate the replacement of the malfunctioning infusion sets by ~2 h on average. Conclusions: On the considered dataset, the proposed algorithm showed the potential to improve the safety of patients treated with sensor-augmented pump systems.

Download Full-text

Dynamic Codebook for Foreground Segmentation in a Video

ECTI Transactions on Computer and Information Technology (ECTI-CIT) ◽

10.37936/ecti-cit.2016102.63738 ◽

2017 ◽

Vol 10 (2) ◽

pp. 144-155

Author(s):

Worapan Kusakunniran ◽

Rawitas Krungkaew

Keyword(s):

Computer Vision ◽

Real World ◽

Early Stage ◽

Image Sequences ◽

Foreground Segmentation ◽

Background Information ◽

Detection Algorithms ◽

Change Detection Algorithms ◽

Novel Method

The foreground segmentation in a video is a way to extract changes in image sequences. It is a key task in an early stage of many applications in the computer vision area. The information of changes in the scene must be segmented before any further analysis could be taken place. However, it remains with difficulties caused by several real-world challenges such as cluttered backgrounds, changes of the illumination, shadows, and long-term scene changes. This paper proposes a novel method, namely a dynamic codebook (DCB), to address such challenges of the dynamic backgrounds. It relies on a dynamic modeling of the background scene. Initially, a codebook is constructed to represent the background information of each pixel over a period of time. Then, a dynamic boundary of the codebook will be made to support variations of the background. The revised codebook will always be adaptive to the new background's environments. This makes the foreground segmentation more robust to the changes of background scene. The proposed method has been evaluated by using the changedetection.net (CDnet) benchmark which is a well-known video dataset for testing change-detection algorithms. The experimental results and comprehensive comparisons have shown a very promising performance of the proposed method.

Download Full-text