Vehicle Detection and Categorisation Using Deep Learning Algorithms

Abstract

Thisresearchaddressesthecrucial task of real-time vehicle detection and traffic analysisonhighwaysandbusyroads,employing advanced object detection algorithms & deep learning techniques. The study focuses on identifyingvariousvehicletypes,includingcars, SUVs, bikes, buses, and trucks, using four distinct algorithms: Single Shot Multibox Detector (SSD), Kalman FilterAlgorithm, You Only Look Once (YOLO v7), and Mask Regional-ConvolutionalNeuralNetwork(Mask R-CNN). The main goal of the research is to applythesealgorithmsinreal-worldsettingson busy metropolitan roads and highways for purposeoftrafficanalysis.Thestudysoftwareis designed to monitor traffic flow and count the numberofcarsthatpassbyinagivenperiodof time, like a day, a week, or a month. Additionally,thealgorithmsortsthesecarsinto various categories and offers a thorough statistical breakdown of the normal vehicle composition on the road under observation. The research\'s conclusions help politicians, transportation engineers, and urban planners by providing insightful information about traffic patterns. These algorithms enable data- drivenanalysis,whichinturninformsdecisions ontrafficmanagement,roadinfrastructure, andsafety protocols. Through the integration of state-of-the-art technology with practical applications, this research makes a substantial contribution to the improvement of traffic monitoring systems, thereby facilitating safer and more intelligent urban movement.

Introduction

Overview

The integration of Artificial Intelligence (AI) in computer vision has revolutionized vehicle detection and traffic analysis, crucial for modern intelligent transportation systems. This review explores the use and comparative performance of advanced algorithms like SSD (Single Shot MultiBox Detector), YOLO (You Only Look Once), Faster R-CNN, Kalman Filter, and Mask R-CNN.

Key Objectives

• Evaluate and compare popular AI algorithms for vehicle detection.

• Examine their performance in real-world traffic conditions.

• Provide insights for improving automated traffic management systems.

Highlighted Algorithms and Findings

1. Single Shot MultiBox Detector (SSD)

• Fast, single-stage object detector suitable for real-time use.

• Achieved up to 74.3% mAP on PASCAL VOC dataset.

• Improved versions use deep feature fusion to increase accuracy (up to 76.3% mAP).

• Effective in detecting both vehicles and pedestrians with over 90–95% detection rate.

2. You Only Look Once (YOLO)

• Known for high-speed and accurate real-time detection.

• Enhanced versions (e.g., YOLOv5) improve small object detection and reduce false positives using techniques like Flip-Mosaic.

• MME-YOLO integrates LiDAR and camera data, reaching 92.8% mAP for robust performance under challenging conditions.

3. Faster R-CNN

• Two-stage detector, offering high accuracy but slower performance.

• Commonly used as a baseline for evaluating object detection models in traffic.

4. Mask R-CNN

• Extends Faster R-CNN with instance segmentation, allowing more precise vehicle localization.

• Used for:

  • Real-time detection.

  • Vehicle type/brand classification.

  • Adaptation to environmental changes (fog, lighting).

  • Remote sensing and geospatial object detection.

5. Kalman Filter

• Used for vehicle tracking in video streams.

• Combines detection with motion prediction.

• Effective but sensitive to noise and occlusion.

Comparative Insights

• SSD and YOLO are best for speed and real-time applications.

• Faster R-CNN and Mask R-CNN provide greater accuracy and segmentation detail, ideal for complex environments.

• Kalman Filter is efficient for tracking but needs enhancements to handle real-world noise.

Recent Innovations

• MME-YOLO: Multi-sensor fusion (LiDAR + camera).

• Improved YOLOv5: Enhanced for small target detection and real-time highway surveillance.

• RES-YOLO: Specialized for remote sensing with up to 93.4% accuracy.

• Flip-Mosaic: Boosts detection of small objects by augmenting training data.

• Instance-based methods: Used for detailed object classification and vehicle counting.

Applications

• Traffic monitoring.

• Autonomous driving.

• Urban planning.

• Remote sensing and surveillance.

• Vehicle classification and speed estimation.

Conclusion

In this comprehensive review, we embarked on a journeythroughthelandscapeofvehicledetection and categorization, leveraging the power of deep learningalgorithms.Ourexplorationencompassed seminalapproachesincludingSSD,MaskR-CNN, YOLO, and the integration of Kalman filtering. Through a meticulous examination of each method, we studied their performance in real- world scenarios. The comparative analysis revealed intriguing insights. SSD showcased commendable speed in detection, making it particularly well-suited for real-time applications. Mask R-CNN, on the other hand, excelled in precise localization, demonstratingitsprowessintasksdemandingfine- grained object delineation.YOLO, with its unique single-shot detection paradigm, struck a balance between accuracy and speed, rendering it a versatile contender in a spectrum of scenarios. The incorporation of Kalman filtering introduced aninvaluabledimensiontotracking,enhancingthe robustness of the algorithms in dynamic environments. Its ability to predict object trajectories and rectify discrepancies brought a temporal coherence to the detections, bolstering the overall performance. Theimplicationsofthesefindingsarefar-reaching. Our insights not only inform the choice of algorithm based on specific application requirements but also pave the wayfor innovative integrations and optimizations. Furthermore, in safety-criticalcontextssuchasautonomousdriving and surveillance, the nuances we uncovered carry profound significance. As we look ahead, this review paper illuminates avenues for further exploration. The synergistic fusion of deep learning with traditional computer vision techniques, the investigation of novel architectures, and the application of these algorithms in multi-modal sensor fusion contexts are promising frontiers. In conclusion, our journey through the realm of vehicle detection and categorization using deep learning algorithms has enriched our understanding of the capabilities and nuances of SSD,MaskR-CNN,YOLO,andtheaugmentative roleofKalmanfiltering.Theseinsightsmarkasubstantialstepforwardinharnessingthepotential ofdeeplearningfortasksofcriticalimportance.As the field continues its rapid evolution, this review servesasbothatestamenttothecurrentstateofthe art and a compass guiding future explorations in this dynamic domain.

References

[1] Kumar, Ashwani, and SonamSrivastava. \"Object detection system based on convolution neural networks using single shot multi-box detector.\"ProcediaComputerScience171 (2020): 2610-2617. [2] Bai, Dongxu, et al. \"Improved single shot multibox detectortargetdetectionmethodbased on deep feature fusion.\"Concurrency and Computation: Practice and Experience34.4 (2022): e6614. [3] Liu,Wei, et al. \"Ssd: Single shot multibox detector.\"Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands,October11–14,2016,Proceedings, Part I 14. Springer International Publishing, 2016. [4] Arinaldi,Ahmad, JakaAryaPradana, and ArlanArventaGurusinga. \"Detection and classification of vehicles for traffic video analytics.\"Procediacomputerscience144 (2018): 259-268. [5] Ojha, Apoorva, SatyaPrakashSahu, and Deepak Kumar Dewangan. \"Vehicle detection through instance segmentation using mask R- CNN for intelligent vehicle system.\" 2021 5th international conference on intelligent computingandcontrolsystems(ICICCS).IEEE, 2021. [6] Xu, Chenchen, et al. \"Fast vehicle and pedestrian detection using improved Mask R- CNN.\" Mathematical Problems in Engineering 2020 (2020): 1-15. [7] Nafi’i, Mohammad Wahyudi, EkoMulyantoYuniarno, and AchmadAffandi. \"Vehicle brands and types detection using mask R-CNN.\" 2019 International Seminar on [8] IntelligentTechnologyandItsApplications (ISITIA). IEEE, 2019. [9] Tahir, Hassam & Khan, Muhammad Shahbaz& Tariq, Muhammad Owais. (2021). PerformanceAnalysisand Comparison ofFasterR-CNN, Mask R-CNN and ResNet50 for the Detection and Counting of Vehicles. 587-594. 10.1109/ICCCIS51004.2021.9397079. [10] Su, Hao, et al. \"Object detection and instancesegmentationinremotesensingimagery basedonprecisemaskR-CNN.\" IGARSS2019- 2019IEEEInternationalGeoscienceandRemote Sensing Symposium. IEEE, 2019 [11] Mahmoud, Amira, et al. \"Object detection using adaptive mask RCNN in optical remote sensing images.\" Int. J. Intell. Eng. Syst 13.1 (2020): 65-76 [12] Dalal, AL-Alimi, et al. \"Mask R-CNN for geospatial object detection.\" International Journal of Information Technology and Computer Science (IJITCS) 12.5 (2020): 63-72 [13] Li,J.,Zhang,F.,&Shi,J.(2023).MME-YOLO:Amulti-modal vehicle detection system using LiDAR and camera data. Sensors, 21(1), 27. [14] Zhang,Y., Guo, Z., Wu, J., Tian,Y., Tang, H., &Guo, X. (2022). Real-time vehicle detection based on improved YOLO v5. Sustainability, 14(19), 12274. [15] Qiu, Y. (2020) Video-Based Vehicle Detection in Intelligent Transportation System. Master Thesis, Jilin University, China. [16] Rodríguez-Rangel, H.; Morales-Rosales, L.A.; Imperial-Rojo, R.; Roman-Garay, M.A.; Peralta-Peñuñuri, G.E.; Lobato-Báez, M. Analysis of Statistical andArtificial Intelligence Algorithms for Real-Time Speed Estimation Based on Vehicle Detection with YOLO. Appl. Sci. 2022, 12, 2907. [17] Uzar,M.,Öztürk,?.,Bayrak,O.C.,Arda,T., &Öcalan, N.T.(2021). Performance analysis of YOLO versions for automatic vehicle detection from UAV images. Advanced Remote Sensing, 1(1), 16-30. [18] WeiLi,Li,Q.L.,&He,J.F.(2022)Vehicle detectioninfoggyweatherbasedonanenhanced YOLO method. [19] Li,Z.,Zhao,Z.,Chen,H.,Zhang,Z.,Xu,Y.,& Liu, Y. (2022). Improved RES-YOLO for Automatic Vehicle Recognition in Vision Measurement and Remote Sensing. Remote Sensing, 22(10), 3783. [20] Moridani, AhadKarimi, SeyyedehHooraFakhrmoosavy, and Mohammad KarimiMoridani. \"Vehicle detention and tracking in roadwaytrafficanalysis usingKalman filter and features.\" International Journal of Imaging and Robotics 15, no. 2 (2015): 45-52. [21] Zhang, Xinyu, HongboGao, Chong Xue, Jianhui Zhao, and Yuchao Liu. \"Real-time vehicle detection and tracking using improved histogram of gradient features and Kalman filters.\"International Journal of Advanced RoboticSystems15,no.1(2018):1729881417749949. [22] Kim, Jeong-ah, Ju-Yeong Sung, and Se- ho Park. \"Comparison of Faster-RCNN,YOLO, and SSD for real-time vehicle typerecognition.\"2020IEEEinternational conferenceonconsumerelectronics-Asia(ICCE- Asia). IEEE, 2020.

Copyright

Copyright © 2025 Sahil Dhumane , Kartik Donwade, Vedant Ghumade, Raghvendra Kalkatuki, Harshada Mhaske. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.