Volume 9, Issue 7, July – 2024 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUL1393

Woodlog Inventory Optimization using Object

Detection and Object Tracking
Vinay Borkar1; Liya T Mathew2; Bhusan Patil3; Bharani Kumar Depuru4
1
Research Associate, AiSpry, Hyderabad, India.
2
Research Associate, AiSpry, Hyderabad, India.
3
Team Leader, Research and Development, AiSpry, Hyderabad, India
4
Director,AiSpry,Hyderabad,India.

Abstract:- In the timber industry, keeping track of wood I. INTRODUCTION

stacks is crucial to avoid stockpiles that are too large or
too small ensuring efficient operations. Traditional Effective inventory management is vital throughout
methods are often time-consuming and inaccurate, numerous industries, which include the timber enterprise.
leading to inefficiencies. This research proposes a novel Improper control in this region often leads to older wood
solution for timber yard inventory management using stacks being used earlier than more recent ones, ensuing in
cutting-edge technology.The approach utilizes YOLOv8, wastage due to decay. This defeats the cause of the usage of
a powerful object detection algorithm, to identify wood clean timber. To prevent this, it's miles critical to have a
logs in live video feeds. DeepSORT, a tracking algorithm, technique to distinguish older timber from more moderen
then follows these identified logs over time. This stacks. The precept of "First-In-First-Out" (FIFO) need to be
automation eliminates the need for manual tracking, carried out, making sure that the primary wooden stack that
minimizes errors, and enforces the "First-In-First-Out" enters is the one used first.
(FIFO) principle for efficient inventory use.The study also
compares DeepSORT to other tracking algorithms like Wooden is an important fabric in production, applied
OC-SORT and ByteTrack to identify the most effective globally for buildings both massive and small. This includes
option for this specific application. The results wooden for building homes of six or more storeys, where
demonstrate that the proposed method significantly know-how the biochemistry and chemistry of wooden change
improves detection accuracy and reliability, leading to can allow even larger structures.[1]
better inventory management.In essence, this research
highlights how integrating advanced detection and Stock control is crucial for almost each sort of business,
tracking technologies can revolutionize timber yard whether service or product-orientated. It affects almost each
operations. By automating processes and ensuring factor of operations, requiring a balance to preserve proper
accurate inventory control, these technologies can stock with minimal economic impact at the consumer.[2]
significantly reduce costs and boost overall efficiency in
the timber sector. FIFO, an acronym for "First-In, First-Out," is a
mechanism for stock manage that organizes stack primarily
Keywords:- Woodlogs Inventory, First-in-First-Out (FIFO), based at the date obtained and sold. The primary batch of
Timber Industry, Object Detection, Object Tracking, merchandise acquired need to be the primary disbursed to
YOLOv8, DeepSORT, ByteTrack, OC-SORT. customers. In inventory accounting models, FIFO is a
perpetual inventory device, ensuring the oldest stack is used
first, preserving stock clean. This method improves material
utilization and decreases loss due to stack expiration.[3]

Our project goals to revolutionize wooden stack control

via assigning a completely unique identifier to each timber
stack, making sure strict adherence to the FIFO precept. This
approach will decorate raw material usage efficiency and
decrease waste within the timber enterprise. The approach to
this research study is based at the CRISP-ML(Q) technique
available as open-supply at the 360DigiTMG internet site.
(ak.1) [Fig. 1]

IJISRT24JUL1393 www.ijisrt.com 2139

Volume 9, Issue 7, July – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUL1393

Fig 1 CRISP-ML (Q) Methodological Framework, Outlining its Key Components and Steps Visually.
Source: - Open-Source ML Workflow Tool- 360DigiTMG

II. LITERATURE REVIEW demand forecasts, thereby improving the overall efficiency of
inventory management processes .[6]
Inventory management is essential to the success of any
business and plays an important role in ensuring profitability. Object tracking, mainly within the context of inventory
Small and medium-sized enterprises (SMEs), which typically management, also benefits from advances in technology. Han
operate on limited budgets, often rely on manual methods of et al. (2004) proposed a more than one speculation approach
inventory management. While this manual process is for monitoring multiple objects, which integrates object
common, it can be labor intensive, error prone, and generally detection and monitoring to ensure a sturdy and efficient
inefficient. The use of an inventory management system has monitoring set of rules . This technique uses a neural
been shown to improve the efficiency and accuracy of these network-based totally item detection module to become
methods.[4] aware of and tune items, offering real-time updates and
feedback for progressed accuracy.[7]
Manual inventory work presents many challenges.
These include time-consuming inventory counts, difficulty in Sun et al. (2021) introduced a Deep Affinity Network
tracking inventory levels, increased risk of error, and (DAN) that learns target object appearances and their
increased costs associated with over-inventory or excess affinities in pairs of video frames. This technique enhances
inventory. Salih et al. (2023) highlighted these problems in item tracking by means of incorporating hierarchical feature
their study on adopting automated inventory management gaining knowledge of, which improves the accuracy of
systems in SMEs, emphasizing the potential for such systems tracking over more than one ranges of abstraction .[8]
to streamline operations and improve accuracy in inventory
tracking .[4] Behrendt et al. (2017) presented a deep learning
approach for real-time searches and tracking with OpenCV.
In an SQL- and PHP-based program implementation of Their pipeline includes steps for detecting, classifying and
an automated catalog management system, Mascarenhas et al. tracking objects, with applications ranging from traffic light
(2020) demonstrated the practical advantages of such detection to complex situations Similarly, Liu et al. (2018)
technologies. Their system was designed to monitor developed an effective fruit first algorithm that combines
inventory in real time, significantly reducing the manual deep learning, tracking, and planning from motion (SfM).
effort required and minimizing the possibility of error.[5] Their system proved to be highly accurate in fruit counting in
image sequences, and demonstrated deep learning in practical
Accurate demand forecasting is another important observational experiments.[9,10]
component of effective inventory management. Inaccurate
expectations can lead to excess inventories or inventories, In the end, the combination of computerized systems
both of which are important.Albayrak et al. (2023) reviewed and superior tracking technology can considerably improve
the applications of artificial intelligence in inventory the efficiency and accuracy of stock control, especially for
management, noting that AI can enhance the precision of SMEs. By decreasing manual effort and minimizing

IJISRT24JUL1393 www.ijisrt.com 2140

Volume 9, Issue 7, July – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUL1393

mistakes, these technologies allow companies to better workflow and present experimental results demonstrating its
manage their sources, forecast demand appropriately, and in effectiveness in enhancing object detection performance.
the end enhance their general operational efficiency.
 Overview of YOLO Algorithm:
III. METHODOLOGY AND TECHNIQUES YOLOv8 is an advanced version of the YOLO object
detection algorithm that uses a deep neural network to
A. Object Detection analyze entire images, identifying objects with class
Object detection is a computer vision technique used to probabilities and bounding boxes. It is faster, more accurate,
identify and locate objects in an image or video. Unlike and better at detecting small objects compared to traditional
simple image recognition, which assigns a single label to an methods. YOLOv8 employs a convolutional neural network
entire image, object detection not only labels objects but also (CNN) to extract features, predict bounding boxes, and refine
draws bounding boxes around them. This provides precise these predictions with offsets, calculating confidence by
information about the location of objects within a given combining class probability and bounding box confidence
scene. For instance, while image recognition might label an score.
entire image as containing a "dog," object detection would
draw individual bounding boxes around each dog in the The YOLO algorithm has significantly impacted
image and label each box accordingly. This allows for computer vision, particularly in real-time object detection,
accurate detection of the location and number of objects due to its speed and efficiency. It is popular in applications
present. like video surveillance, self-driving cars, and augmented
reality. One of its main advantages is the ability to process
To leverage the capabilities of object detection images quickly, enabling real-time detection with relatively
techniques like YOLO (You Only Look Once) [11] and little training data.
address its limitations, we propose a new workflow that
combines YOLO's efficiency with additional methods [12] However, YOLO also has limitations. It may struggle
(ak.2) Fig 2 . This workflow involves a multi-step approach with accurately detecting small objects and performing fine-
where YOLO's initial detections are refined using grained classification. Despite these challenges, YOLO has
sophisticated post-processing algorithms. Through feature- advanced object detection and created new opportunities for
based refinement and context analysis, our proposed computer vision research and applications. As the field
workflow aims to improve detection accuracy, particularly evolves, improvements to YOLO and other algorithms will
for scenarios involving small objects or fine classification. In likely address these limitations and meet new challenges.
the following sections, we detail the components of this

Fig 2 ML Workflow Architecture: A comprehensive overview of the ML pipeline for Wood Logs Inventory Optimization.
Source: Open-Source ML Workflow Tool- 360DigiTMG

B. Object Tracking Real-time Tracking (DeepSORT), OC-SORT, and ByteTrack

Object tracking involves identifying an object in a video combine deep learning with traditional tracking techniques to
and monitoring its movement over time, with applications in achieve state-of-the-art performance.
surveillance, robotics, and self-driving cars. This task is
particularly challenging in complex environments where  Overview of DeepSORT:
objects can become occluded or their appearance can change The Deep Simple Online and Real-time Tracking
due to lighting variations. Traditional algorithms often rely (DeepSORT) algorithm[13] is an advanced method for
on motion models and feature extraction techniques. In tracking multiple objects simultaneously by associating
contrast, modern methods such as Deep Simple Online and detections across frames. An extension of the SORT (Simple

IJISRT24JUL1393 www.ijisrt.com 2141

Volume 9, Issue 7, July – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUL1393

Online Real Time Tracking) algorithm, DeepSORT adds a using a combination of the Intersection over Union (IOU)
deep learning-based appearance descriptor to reduce identity metric and the appearance descriptor. The Kalman filter’s
switches and enhance tracking efficiency. DeepSORT uses state is then updated using the associated detection, and the
the Kalman filter to predict an object's location in the next process is repeated for the subsequent frames. Fig 3 shows
frame and associates the detection with the predicted location the overview of the DeepSORT algorithm.

Fig 3 : Overview of the DeepSORT algorithm

Source: ref. [14]

 Overview of ByteTrack: understanding of the project architecture. Here’s how it

ByteTrack[15] represents a significant advancement in works:
multi-object tracking by incorporating all detection boxes,
including those with low confidence scores, into the  Warehouse Inventory: Wood logs are stored and managed
association process. This approach results in more accurate based on the FIFO principle.
and continuous tracking, particularly in complex and  Cameras: Multiple cameras capture images or video
dynamic environments. The algorithm's performance on frames of wood log stacks from various angles in the
benchmark datasets underscores its effectiveness and warehouse.
potential for real-world applications.  Computing Resources: A powerful computer or server is
used for training and deploying the Object detection and
 Overview of OC-SORT: object tracking model, depending on dataset size and
OC-SORT[16] represents a significant evolution of the complexity.
SORT algorithm by adopting an observation-centric  Dataset Collection: A diverse dataset of wood log stack
approach that prioritizes accurate data association and robust images or frames is collected to train the models, ensuring
handling of occlusions and appearance changes. This results coverage of different positions, orientations, and lighting
in improved multi-object tracking performance, making it conditions.
highly effective for various real-world scenarios where  Annotation Tools: Tools like Roboflow label the dataset
traditional motion model-based tracking methods may fall by drawing bounding boxes around stacks and assigning
short. class labels.
 Tracking Algorithm: After training, Object tracking
C. Proposed Method for Wood Logs Inventory Optimization: algorithms track wood log stacks in real-time, assigning
This paper introduces a method for detecting and unique identifiers for accurate inventory management. It
tracking wood log stacks in a warehouse using the object uses appearance features and motion information to
detection and tracking algorithms. Ref. [Fig 4] for better ensure precise tracking.

IJISRT24JUL1393 www.ijisrt.com 2142

Volume 9, Issue 7, July – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUL1393

Fig 4 Model Architecture

D. Implementation Steps:  Data Augmentation (Flip, Rotation, Shear, Hue,

Figure 5 depicts the Project Implementation Roadmap, Saturation, Brightness, Noise)
which will be detailed in the subsequent sections. Data enhancement techniques such as flipping, rotation,
shearing, hue adjustment, saturation, brightness and adding
 Extracting Images from Videos : noise are applied to the annotated images to enhance the
The initial step involves extracting frames from video robustness and generalizability of the model.
recordings of wood log stacks. This process ensures a diverse
and comprehensive dataset capturing various angles and  Train Test Split
conditions of the wood log stacks. The data set is divided into training and testing subsets.
This classification allows the performance of the model to be
 Frame Extraction, Frame Selection, Frame Resolution & checked on unseen data, thus improving reliability and
Quality accuracy.
Once the frames are picked, we carefully select the ones
with the desired shape and quality. This selection process  Loading Dataset to YOLO Model for Training
ensures that the next step works on the most informative and A training subset of the annotated dataset is then loaded
high-quality images. into the YOLOv8 model. This phase involves configuring the
model with appropriate parameters and initializing the
 Reading Frames and Labeling Using Roboflow training process.
The selected frame is read into our system and then
encoded by Roboflow. In this step, each frame is annotated  Loading the Trained Model for Detection
with a bounding box, and class labels are provided to identify Upon completion of the training process, the trained
the different wooden log stacks. YOLOv8 model is loaded to detect wood log stacks. This step
involves testing the model on various scenarios to ensure it
 Annotation Tools (Bounding Box Annotation, Class can accurately identify and classify the log stacks.
Labels)
Annotation tools are used to create bounding boxes  Tracking Algorithms (Deep SORT, OC-SORT, Byte
around each wood-log stack and assign appropriate class Track)
labels. This labeled data set is important for training the To facilitate continuous monitoring and tracking of the
analytical model. wood log stacks, the trained detection model is integrated
with tracking algorithms such as Deep SORT, OC-SORT,
and BYTE TRACK.These algorithms maintain the identity of
each log stack over time , even in crowded places.

IJISRT24JUL1393 www.ijisrt.com 2143

Volume 9, Issue 7, July – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUL1393

 Model Evaluation (MOTA, MOTP, IDF)  Desired Output where Stacks are Labeled Based on When
The model's performance is evaluated using metrics They Come into the Frame Obtained
such as Multiple Object Tracking Accuracy (MOTA), The system tracks and registers the stacks of logs based
Multiple Object Tracking Precision (MOTP), and Identity F1 on the time they enter the frame, following FIFO (First-In-
Score (IDF). These metrics assess the model's detection and First-Out) principles. The combination of the YOLOv8
tracking capabilities, ensuring it meets the required standards. model with Deep SORT ensures that each stack is constantly
monitored and accurately identified. As the characters enter
 Best Model Selected and DeepSORT Tracker Updated the frame, they are assigned unique signals and their positions
with the IDS and Bounding Boxes are updated in real time. This approach ensures that the
Based on the results of the analysis, the best performing catalog is maintained, and that old records are processed
model was selected. The selected model is then merged with before new ones are created, so key.
the Deep SORT tracker, which updates the IDs and bounding
boxes for each known log stack.  Deployment
The fully trained and validated model, integrated with a
 Bounding Boxes Along with Their Classes Shown tracking algorithm, has been successfully deployed on the
The system displays bounding boxes with their class Streamlit platform for real-time monitoring. This deployment
labels around each detected tree log stack. This visual image currently focuses on monitoring wood log stacks using an AI-
helps verify the search and tracking process. driven camera system. Future work includes extending this
deployment to a warehouse environment for broader
application.

Fig 5 Project Implementation Roadmap

 Data Collection  Here's the Process:

For this research project, the dataset consists of 497
images of stacks. These images are used to develop and train  Data Extraction and Annotation:
a model that identifies and tracks each stack. The model Frames were extracted from sample videos using
assigns a unique identifier to each stack, enabling accurate Python/Roboflow. Each stack case in a frame was labeled
tracking and management. When a stack is removed from the using the Roboflow algorithm, resulting in 497 labeled
frame, the system re-assigns IDs to maintain updated images. Fig 6 shows the bounding boxes drawn by using
inventory records effectively. By doing this we will get to roboflow.
know which stack should be shipped out next from the
inventory.  Annotation Format:
Each image is accompanied by a .txt file containing
 Data Preprocessing bounding box annotations in the format:” class_id center_x
Wood log pile image analysis is crucial for stockpile center_y width height ” [17] Coordinates are normalized from
identification and shipping decisions. Before feeding images zero to one.
into deep learning models, a robust preprocessing pipeline is
essential to enhance data quality, reduce noise, and optimize
model performance.

IJISRT24JUL1393 www.ijisrt.com 2144

Volume 9, Issue 7, July – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUL1393

 Data Augmentation: transformations like translation, cropping, rotation, and color

To increase dataset diversity, data augmentation distortions. Augmentation helps in training robust deep
techniques were applied. These include random learning models by exposing them to varied scenarios.

Fig 6 Views of Object Detection Bounding Boxes, Contributing to a Comprehensive Understanding of Detection Accuracy.

Roboflow[17] is a computer vision tool used by  Shear: ±10° Horizontal, ±10° Vertical,
developers for various deep learning tasks, including  Hue: Between -15° and +15°,
annotation, model building, and deployment. Roboflow  Saturation: Between -25% and +25%,
Annotate is an online tool for labeling images for object  Brightness: Between -15% and +15%,
detection, classification, and segmentation. For this project,  Noise: Up to 0.1% of pixels,
Roboflow was used for annotation and pre-processing.
During the pre-processing stage, we applied auto-orientation, Ref. [Fig: 7, 8, 9, 10, 11, 12] These figures collectively
auto-resizing, and automatic contrast adjustment to the provide a comprehensive understanding of both the dataset
images. In the augmentation process, we applied: and the model's performance. They cover aspects from data
composition and annotation patterns to the model's training
 Flip: Horizontal,Vertical progression and final evaluation metrics, offering a holistic
 Rotate: Clockwise, Counter-Clockwise, Upside Down view of the entire process from data preparation to model
 Rotation: Between -10° and +10°, validation and testing.

Fig 7 Dataset's Images, Annotations, and Average Image Sizes, Offering Insight into its Composition and Characteristics

Fig 8 Distribution of Image Sizes within our Dataset, Highlighting Variations in Dimensions Across the Data.

IJISRT24JUL1393 www.ijisrt.com 2145

Volume 9, Issue 7, July – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUL1393

Fig 9 Annotation Heatmap, Providing a Visual Representation of the Density of Object Annotations within Images

Fig 10 Histogram Illustrating how the Number of Objects per Image is Distributed Across the Dataset

Fig 11 Validation Set Results for Average Precision, Compactly Conveying the Model's Performance.

IJISRT24JUL1393 www.ijisrt.com 2146

Volume 9, Issue 7, July – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUL1393

Fig 12 Test Set Results for Average Precision are Depicted, Showcasing the Model's Performance.

 Model Selection started by carefully selecting a wide range of datasets,

In this project, we explored various object tracking carefully annotating them, and preprocessing the data. To
methods, including DeepSORT, OC-SORT, and ByteTrack. achieve precise stack identification, YOLOv8 configuration
Our approach incorporated a combination of advanced post- required parameter selection and training process
processing and optimization techniques to address challenges optimization. Before combining YOLOv8 with tracking
specific to YOLO, as illustrated in Figure 2 of our Machine algorithms for robust object tracking, we verified its
Learning Workflow Architecture. By evaluating these performance using metrics like mean Average Precision
methods, we aimed to identify the most effective tracking and (mAP) and precision. To make sure our integrated model
recognition techniques. Based on the experimental results satisfies strict requirements for warehouse inventory
from these evaluations, we will select the model that delivers management, it underwent extensive testing in both simulated
the highest performance in terms of detection and tracking and real-world settings before being deployed. Keeping an
accuracy. This approach ensures that we choose the most eye on things and making adjustments as needed are crucial
effective solution for our specific project needs. to keeping our project running well.

 Model Training  YOLO Performance Summary:

Our project trains, deploys, and tests YOLOv8 with the For object detection, we used YOLOv8, achieving the
goal of improving wood log stack recognition and tracking in following performance metrics: Ref. Fig [13] and Fig [14].
warehouses. To achieve the best model performance, we

Fig 13 Training Graphs for the YOLO Model, Presenting its Learning Progress and Performance.

IJISRT24JUL1393 www.ijisrt.com 2147

Volume 9, Issue 7, July – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUL1393

Fig 14 The Reduction of Object Detection Precision Losses via Improved Labeling of Boxes, Classes, and Objects

 Model Evaluation  Precision & Recall: Indicate accuracy of true versus

To effectively evaluate and select the best model among false detections.
tracking algorithms like DeepSORT, OC-SORT, and  MT, PT, ML: Classify tracking completeness of
ByteTrack, several key metrics are employed: trajectories.

 MOTA: Evaluates tracking accuracy, considering All parameters together provide a comprehensive level
false positives, false negatives, and identity switches. of evaluation ,to compare DeepSORT, OC-SORT and
 MOTP: Measures average error between predicted ByteTrack. For detailed investigation and description, tools
and actual positions. such as statistical indicators and evaluation methods on
 IDF1: Balances precision and recall for identity websites such as MOTChallenge [18] provide guidelines for
tracking. evaluating and selecting the best monitoring methods for
 IDP & IDR: Assess accuracy of identity assignments. specific projects.
 FP & FN: Count incorrect and missed detections.
 IDSW: Tracks changes in object identities. These metrics provide a comprehensive evaluation of a
 Track Fragmentation: Measures trajectory tracking algorithm's performance, balancing accuracy,
interruptions. precision, and consistency.

Table 1 Evaluation Metrics of DeepSORT

DeepSORT
MOTA: 0.50
MOTP: 0.75
IDF1: 0.76
IDP: 1.00
Precision: 0.75
Recall: 0.76
Track Fragmentation: 0.24
MT: 0.76
PT: 0.00
Mostly Lost: 0.24

IJISRT24JUL1393 www.ijisrt.com 2148

 Volume 9, Issue 7, July – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUL1393

Given these metrics in table 1, DeepSORT performs the to varying environmental conditions and logistical
best among the three algorithms. It has the highest MOTA, intricacies, holds significant promise. Such research
MOTP, IDF1, IDP, Precision, Recall, and MT, indicating endeavors could yield valuable insights by testing these
better overall tracking performance. systems across different warehouse settings and product
types.
 Model Deployment
We deployed our model using a Streamlit application  Energy Efficiency:
designed for video processing and displaying annotated Given the computational demands of these algorithms,
outputs. The application identifies and labels wood log stacks exploring avenues to enhance their energy efficiency assumes
as they appear in the video stream, numbering each stack heightened relevance, particularly in resource-constrained
based on its appearance order for easy identification and operational environments.
management. Users upload warehouse videos via the
Streamlit interface, where the application processes them  Robustness and Error Handling:
frame by frame using YOLOv8 for object detection and Further research efforts could concentrate on fortifying
DeepSORT algorithm for object tracking. The application the robustness of these systems, ensuring robust error-
provides real-time annotations, displaying each detected and handling mechanisms that effectively manage challenges
tracked stack with a unique identifier. The annotated video such as occlusions, fluctuating lighting conditions, and other
output serves as a visual guide for inventory management, real-world complexities.
ensuring accurate tracking and retrieval of stacks for
shipment. The user-friendly Streamlit interface allows users In essence, addressing these research avenues promises
to upload videos and observe real-time annotations, to advance the efficacy and applicability of advanced object
enhancing operational efficiency and decision-making. This detection and tracking technologies in optimizing wood log
deployment approach optimizes inventory management by inventory management, thereby driving efficiencies and
leveraging advanced detection and tracking technologies to operational enhancements in relevant industrial sectors.
streamline warehouse operations.
V. RESULTS AND DISCUSSION
IV. IMPLICATIONS FOR FUTURE RESEARCH
This examine investigated the impact of mixing
The findings from this study open up several avenues YOLOv8 for item discovery and DeepSORT for tracking to
for future research in the domain of wood log inventory optimize tree log stock control in warehouse structures We
optimization leveraging advanced object detection and accomplished comparative evaluation with different tracking
tracking algorithms: algorithms, together with OC-sort and ByteTrack. Time was
spent focusing on key features which include a couple of item
 Algorithm Improvement: tracking Accuracy (MOTA), more than one object tracking
While YOLOv8 and DeepSORT have exhibited Precision (MOTP), and IDF1 ratings.Our outcomes
promising outcomes, further refinement and evolution of consistently display that the YOLOv8-DeepSORT aggregate
these algorithms stand to augment their precision and outperforms the alternatives,with a excessive diploma of
efficiency. Exploring novel models or hybrid approaches accuracy and reliability required for an powerful inventory
amalgamating the strengths of diverse algorithms could control.
potentially enhance overall performance.
DeepSORT's performance advantage over OC-kind and
 Real-time Processing: ByteTrack highlights its robustness in handling complicated
The imperative focus on implementing and fine-tuning monitoring eventualities in dynamic warehouse
real-time processing capabilities for extensive industrial environments.Utilizing YOLOv8 for unique item detection,
applications remains paramount. Ensuring seamless handling our integrated technique now not handiest boosts monitoring
of substantial data volumes in real-time, without accuracy but additionally complements real-time processing
compromising accuracy, emerges as a critical requirement for abilties. That is vital for adapting to variable lighting
practical deployment scenarios. situations, occlusions, and various warehouse layouts, as
evidenced via the robustness visible within the screenshots
 Integration with IoT: from our video outputs (Ref [Fig 16]).
Future investigations may delve into the integration of
these algorithms with Internet of Things (IoT) infrastructure, The realistic implications of our findings are sizeable for
aimed at establishing a more interconnected and automated advancing inventory control systems in industrial settings.
system for inventory management. This could encompass Integrating superior object detection and monitoring era like
leveraging sensors and smart devices to furnish real-time YOLOv8 and DeepSORT can streamline warehouse
inputs to detection and tracking algorithms. operations, reduce mistakes related to guide tracking, and
optimize useful resource allocation. The tested effectiveness
 Scalability and Adaptability: of our approach underscores its capability to beautify
Evaluating the scalability of these systems across operational performance and accuracy in warehouse logistics.
diverse inventory sizes and types, alongside their adaptability

IJISRT24JUL1393 www.ijisrt.com 2149

Volume 9, Issue 7, July – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUL1393

Fig 15 Output Tracking Results using Deep SORT, OC-SORT and ByteTrack with Bounding Boxes on the Object.

VI. CONCLUSION [5]. Melancy Mascarenhas, Anand Lamani, Chandrakant

Matkar, Amogh Ramchandra Dessai, Anush
In conclusion, this study highlights the potential of Kotharkar . An Automated Inventory Management
comprehensive tracing and tracking systems in the quality System. International Journal of Computer
management of wood relics. By using YOLOv8 for searching Applications. 176, 14 ( Apr 2020), 21-23.
and DeepSORT for tracking, the study achieved remarkable DOI=10.5120/ijca2020920064
improvements in tracking accuracy and efficiency. The [6]. Albayrak Ünal, Özge & Erkayman, Burak &
application of this technology in a real warehouse Usanmaz, Bilal. (2023). Applications of Artificial
demonstrated its practical value for inventory. Future Intelligence in Inventory Management: A Systematic
research should extend these findings by investigating further Review of Literature. Archives of Computational
algorithmic improvements, real-time processing capabilities, Methods in Engineering. 30. 10.1007/s11831-022-
IoT integration, scalability, energy efficiency, and energy 09879-5.
efficiency to further improve the system and they are used in [7]. Mei Han, A. Sethi, Wei Hua and Yihong Gong, "A
a variety of industrial situations. detection-based multiple object tracking method,"
2004 International Conference on Image Processing,
REFERENCES 2004. ICIP '04., Singapore, 2004, pp. 3065-3068 Vol.
5, doi: 10.1109/ICIP.2004.1421760.
[1]. Ramage, Michael & Burridge, Henry & Wicher, [8]. S. Sun, N. Akhtar, H. Song, A. Mian and M. Shah,
Marta & Fereday, George & Reynolds, Thomas & "Deep Affinity Network for Multiple Object
Shah, Darshil & Wu, Guanglu & Yu, Li & Fleming, Tracking," in IEEE Transactions on Pattern Analysis
Patrick & Densley Tingley, Danielle & Allwood, and Machine Intelligence, vol. 43, no. 1, pp. 104-119,
Julian & Dupree, Paul & Linden, P. & Scherman, 1 Jan. 2021, doi: 10.1109/TPAMI.2019.2929520.
Oren. (2016). The wood from the trees: The use of [9]. K. Behrendt, L. Novak and R. Botros, "A deep
timber in construction. 10.17863/CAM.6411. learning approach to traffic lights: Detection,
[2]. Pandey, Deepa & Raut, Niyati. (2016). INVENTORY Ttracking, and classification," 2017 IEEE
MANAGEMENT BY USING FIFO SYSTEM. International Conference on Robotics and Automation
[3]. Ramdasi, Shubhankar & Shinde, Dattaji. (2021). (ICRA), Singapore, 2017, pp. 1370-1377, doi:
Effect of FIFO Strategy Implementation on 10.1109/ICRA.2017.7989163.
Warehouse Inventory Management in The Furniture [10]. "Robust Fruit Counting: Combining Deep Learning,
Manufacturing Industry. International Journal of Tracking, and Structure from Motion," 2018
Engineering and Technical Research. 10. 179-183. IEEE/RSJ International Conference on Intelligent
[4]. Salih, Hayder & G. Yaseen, Mohanad & Aljanabi, Robots and Systems (IROS), Madrid, Spain, 2018, pp.
Mohammad. (2023). Implementing an Automated 1045-1052, doi: 10.1109/IROS.2018.8594239.
Inventory Management System for Small and [11]. Terven, J.; Córdova-Esparza, D.-M.; Romero-
Medium-sized Enterprises. Iraqi Journal for Computer González, J.-A. A Comprehensive Review of YOLO
Science and Mathematics. 4. 238-244. Architectures in Computer Vision: From YOLOv1 to
10.52866/ijcsm.2023.02.02.021. YOLOv8 and YOLO-NAS. Mach. Learn. Knowl.
Extr. 2023, 5, 1680-1716.
https://doi.org/10.3390/make5040083

IJISRT24JUL1393 www.ijisrt.com 2150

Volume 9, Issue 7, July – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUL1393

[12]. URL : Machine Learning Workflow Architecture for

Beginners - 360DigiTMG
[13]. N. Wojke, A. Bewley and D. Paulus, "Simple online
and realtime tracking with a deep association metric,"
2017 IEEE International Conference on Image
Processing (ICIP), Beijing, China, 2017, pp. 3645-
3649, doi: 10.1109/ICIP.2017.8296962. keywords:
{Kalman filters;Tracking;Extraterrestrial
measurements;Standards;Uncertainty;Cameras;Comp
uter Vision;Multiple Object Tracking;Data
Association},
[14]. Singh, I.S., Wijegunawardana, I.D., Samarakoon,
S.M.B.P. et al. Vision-based dirt distribution mapping
using deep learning. Sci Rep 13, 12741 (2023).
https://doi.org/10.1038/s41598-023-38538-3
[15]. Zhang, Y. et al. (2022). ByteTrack: Multi-object
Tracking by Associating Every Detection Box. In:
Avidan, S., Brostow, G., Cissé, M., Farinella, G.M.,
Hassner, T. (eds) Computer Vision – ECCV 2022.
ECCV 2022. Lecture Notes in Computer Science, vol
13682. Springer, Cham. https://doi.org/10.1007/978-
3-031-20047-2_1
[16]. J. Cao, J. Pang, X. Weng, R. Khirodkar and K. Kitani,
"Observation-Centric SORT: Rethinking SORT for
Robust Multi-Object Tracking," 2023 IEEE/CVF
Conference on Computer Vision and Pattern
Recognition (CVPR), Vancouver, BC, Canada, 2023,
pp. 9686-9696, doi:
10.1109/CVPR52729.2023.00934.
[17]. Dwyer, B., Nelson, J., Hansen, T., et. al. (2024).
Roboflow (Version 1.0) [Software]. Available from
https://roboflow.com. computer vision.
[18]. Dendorfer, Patrick & Os̆ep, Aljos̆a & Milan, Anton &
Cremers, Daniel & Reid, Ian & Roth, Stefan & Leal-
Taixé, Laura. (2021). MOTChallenge: A Benchmark
for Single-Camera Multiple Target Tracking.
International Journal of Computer Vision.

IJISRT24JUL1393 www.ijisrt.com 2151