Traffic Sign Board Recognition and Voice Alert System using Convolution Neural Network

DOI : 10.17577/IJERTCONV10IS12023

Download Full-Text PDF Cite this Publication

Text Only Version

Traffic Sign Board Recognition and Voice Alert System using Convolution Neural Network

Sampada P S, Shakeela A, Simran Singh, Supriya J, Kavya M Students, CSE Department, Sri Krishna Institute of Technology, Blore-560090, India Faculty, CSE Department, Sri Krishna Institute of Technology, Blore-560090, India

Abstract:-To ensure smooth and safe circulation, road signs are essential. A major cause of traffic accidents is neglect in posting and misinterpreting signage. The proposed system assists in identifying the road sign and sending a voice alert to the driver through the speaker so that he/she can make the necessary decisions. The proposed system is formed using the convolutional neural network (CNN), which helps in the recognition and classification of traffic signage images. A set of classes is defined and formed into a specific data set to make it more precise. The German Traffic Sign Benchmarks Dataset was utilized, which contains around 43 categories and 51,900 pictures of traffic signs. The accuracy of the execution is almost 98.52 percent. Taking after the location of the sign by the system, a voice caution is sent through the speaker which informs the driver. The proposed system too contains an area where the vehicle driver is alarmed around the traffic signs in the near vicinity which makes a difference them to be mindful of what rules to take after on the course. The aim of this system is to guarantee the security of the vehicle's driver, travelers, and pedestrians.

KeywordsConvolutional Neural Network, GTSRB Dataset, Traffic Signs, Voice Alert


    Traffic signs location and acknowledgment is an critical viewpoint for giving security for all the street clients. Various computer vision frameworks have been created as of late for traffic signs examination. But the characteristics of existing calculations (acknowledgment accuracy, false caution rate, vigor against air changes) are still not great sufficient to substitute a human operator. Over the world, 1.2 million individuals were slaughtered in traffic crashes in 2002, which was 2.1% of all worldwide passings and the 11th positioned cause of passing [1]. These days, we are moving towards a unused period in which, much obliged to advances, crashes are uncommon or maybe then commonplace. In truth, modern Brilliantly Transport Frameworks (ITS) have been presented in automotive industry in arrange to spare cash and lives, and to create the driving secure and convenient. There have been a part of innovative headways and cars with auto-pilot mode have come up. Independent vehicles have come into presence. There has been a boom in the self-driving car industry. Be that as it may, these highlights are accessible as it were in a few high-end cars which are not reasonable to the masses. We needed to plan a framework which makes a difference in easing the work of driving to a few extent. On conducting a overview, we found that the size of street mischances in India is alarming. Reports recommend that each hour there are approximately 53 disasters taking put on the roads. Additionally, each hour more than 16 passings happen due to these disasters. When someone neglects to comply traffic signs whereas driving, they are putting their life as well the life of the other drivers, their

    travelers and those on the street at hazard. Consequently, we came up with this system in which traffic signs are naturally identified utilizing the live video stream and are read out out loud to the driver who may at that point take the specified choice. Another range of center in our framework is the thought of getting the area of the client utilizing GPS. Moreover, all the traffic signs will be put away in a database together with their area so that the driver will be informed in advance with respect to the another drawing nearer traffic Sign.


    In this period of a fast-paced life, individuals for the most part tend to miss out on recognizing the traffic sign and subsequently break the rules. A part of inquire about has been wiped out this space in arrange to reduce the number of mischances. Analysts have utilized a assortment of classification algorithms and a number of CNN models to classify the traffic signs and alarm the driver. Our system points to optimize the method of acknowledgment and at the same time give other benefits such as early caution to the driver. The location of traffic signs has been exhausted a assortment of procedures in various studies. [1] One of the forms utilizes the Bolster Vector Machine strategy. The dataset was divided into 90/10 for preparing and testing purposes, and it utilizes straight classification. To achieve the required result, a arrangement of stages called Color Division, Shape Classification, and Acknowledgment were taken after. Raspberry Pi is utilized in identifying and recognizing traffic signs with much less coding [2]. However, it requires the Raspberry Pi board at one's talk for usage which is quite costly. Another way of traffic sign acknowledgment is picture seriously [3]. A video is procured and broken down into outlines. Picture pre-processing is done which incorporates isolating the foreground and the foundation, diminishing and differentiate upgrade. The signs are then categorised as hexagonal, triangular, or circular in shape and transmitted for template matching after these operations. The objects with a few positive shapes are coordinated from the pretrained algorithm.

    Caffe, an open-source framework, that makes a difference to identify and perceive street traffic signs with high accuracy and effectiveness [4]. A CNN approach is proposed for preparing traffic sign training sets and getting a demonstrate that can categorise traffic signs. Another strategy for utilizing the CNN conspire is proposed in [11], in which the real border of the objective sign is evaluated by projecting the boundary of a comparing format sign picture into the input picture plane. The strategy propels to gotten to be end-to-end trainable when we change the boundary estimation issue into a posture and shape expectation work based on CNN. It is more safe to occlusion and limit goals than other boundary assessing methods that

    center on contour estimation or picture segmentation.

    [6] proposes a multi-resolution include combination arrange engineering for sign discovery, which aids within the partition of various little objects from sign sheets. A vertical spatial sequence attention (VSSA) module can too be utilized to assemble additional setting data for improved detection. Utilizing GPS-based following, Expanded Reality innovation is joined in mobile apps [5]. It employments the facilitates of a user's smartphone as a pointer to help people in dynamically and basically finding conceivable assets within the quick region based on the direction of the user's camera see.

    In [7] AlexNet structure of CNN is utilized in which the design contains eight layers. The first five layers are convolutional layers and the last mentioned three are all associated layers. The accuracy of this design comes out to be 92.63%. Too, the GoogleNet design is implemented in [7] which makes a difference in working with huge information and a tall number of parameters. However, it comes with an issue that the huge information causes organize overfitting diminishing the accuracy to fair 80.5%. VGG CNN is proposed in [8] which features a essentially superior performance as compared to other accessible designs. The number of parameters in this approach is impressively decreased in arrange to enhance and speed up the calculation. The network to incorporates the BN (clump standardization) and Crevice (worldwide normal pooling) layers, which offer assistance to progress precision without expanding the number of parameters. However, we discovered in [10] that by erasing the pool4 layer of VGG16 and utilizing enlargement for ResNet, we can combine the progressed design of Faster-RCNN with Online Difficult Examples Mining (OHEM), making the framework more flexible and helping within the discovery of minor traffic signs.


    Existing System: We discover nearly a significance of all the over papers in because it presents a mini-batch proposal selection instrument in combination with a profound various leveled design that permits the neural organize to distinguish the traffic signs as well as the traffic lights by preparing them on separate datasets. The strategy settle the issue of occurrences from one dataset not being labelled within the other. The framework makes a difference in giving a modern measurement to our venture by providing the thought of traffic sign localization for driver help.

    Proposed System:

    Dataset In the proposed framework, the German Traffic Sign Benchmarks (GTSRB) Dataset is employments with 43 distinctive traffic signs that are considered to prepare the demonstrate.

    Information Pre-processing :

    To perform picture handling, pictures have to be changed over into NumPy clusters (i.e. numeric values). After stacking the pictures, they are resized to 30*30 pixels. Post this, the names of the image are mapped with the picture and thus the dataset is prepared to be prepared.

    Model :

    Convolutional Neural Arrange (CNN) is an calculation falling within the space of Deep Learning. CNN can take a picture as input, dole out priority to distinctive things within the picture, and recognize them from one another. It requires much less pre-

    processing as compared to other classification calculations. Convolutional Organize has the capacity to memorize the channels or characteristics within the pictures as restricted to the primitive strategies channels where they are done manually. Below figure 1 speaks to the number of layers utilized within the model. There are 4 convolution layers and 2 max pooling layers along side dropout, smooth and thick layers. Adam optimizer is utilized within the neural arrange. The input estimate of the picture is 30*30*1. The demonstrate utilizes the ReLU actuation work. We get a completely associated layer after the Smooth layer. and finally, the yield is decided by utilizing the softmax enactment work.

    Figure 1. Neural Network Representation


    To ensure a smooth and secure flow of traffic, road signs are essential. A major cause of road accidents is negligence in viewing the Traffic signboards and interpreting them incorrectly. A system that can help in recognizing the Traffic sign and sending a voice alert through the speaker to the driver so that he/ she may take necessary decisions. Traffic sign detection and recognition plays an important role in expert systems, such as traffic assistance driving systems and automatic driving systems. It instantly assists drivers or automatic driving systems in detecting and recognizing traffic signs effectively. The traffic sign board recognition focuses on reduction of the traffic load on existing road network through various travel demand management measures.


    The trained neural network which contains 4 convolution layers and 2 max pooling layers along with dropout, flatten and dense layers proved to give a better result as compared to the other CNN Architectures. The accuracy of the trained network is 98.52%.


    The Traffic Sign Board Recognition and Voice Alarm System is executed utilizing Convolutional Neural Network. Various models beneath the CNN heading were examined and the one with highest precision on the GTSRB dataset was implemented. The creation of diverse classes for each Traffic sign has helped in expanding the precision of the show. A voice message is sent after acknowledgment of the sign which cautions the driver. A outline is shown on which the signs within the region of the driver are shown hence making a difference him/her take appropriate decisions. This paper may be a critical headway within the field of driving because it would ease the work of the driver without compromising on the security perspective. Too this system can easily be actualized

    without the require of much equipment thus increasing its reach.


We would like to thank Assistant Professor Kavya M for her valuable suggestion, expert advice and moral support in the process of preparing this paper.


[1] V. Ciuntu and H. Ferdowsi, "Real-Time Traffic Sign Detection and Classification Using Machine Learning and Optical Character Recognition," 2020 IEEE International Conference on Electro Information Technology (EIT), 2020, pp. 480- 486, doi: 10.1109/EIT48999.2020.9208309.

[2] M. T. Islam, "Traffic sign detection and recognition based on convolutional neural networks," 2019 International Conference on Advances in Computing, Communication and Control (ICAC3), 2019, pp. 1-6, doi: 10.1109/ICAC347590.2019.9036784.

[3] Yadav, Shubham & Patwa, Anuj & Rane, Saiprasad & Narvekar, Chhaya. (2019). Indian Traffic Sign Board Recognition and Driver Alert System Using Machine Learning. International Journal of Applied Sciences and Smart Technologies. 1. 1-10. 10.24071/ijasst.v1i1.1843.

[4] Anushree.A., S., Kumar, H., Iram, I., & Divyam, K. (2019). Automatic Signboard Detection System by the Vehicles.

[5] S. Harini, V. Abhiram, R. Hegde, B. D. D. Samarth, S. A. Shreyas and

K. H. Gowranga, "A smart driver alert system for vehicle traffic using image detection and recognition technique," 2017 2nd IEEE International Conference on Recent Trends in Electronics, Information & Communication Technology (RTEICT), Bangalore, India,2017,pp.15401543,doi:10.1109/RTEICT.2017.8256856.

[6] C. Wang, "Research and Application of Traffic Sign Detection and Recognition Based on Deep Learning," 2018 International Conference on Robots & Intelligent System (ICRIS), Changsha, China, 2018, pp. 150- 152, doi: 10.1109/ICRIS.2018.00047.

[7] M A Muchtar et al 2017 J. Phys.: Conf. Ser. 801 012010

[8] Y. Yuan, Z. Xiong and Q. Wang, "VSSA-NET: Vertical Spatial Sequence Attention Network for Traffic Sign Detection," in IEEE Transactions on Image Processing, vol. 28, no. 7, pp. 3423-3434, July2019, doi: 10.1109/TIP.2019.2896952.

[9] S. Huang, H. Lin and C. Chang, "An in-car camera system for traffic sign detection and recognition," 2017 Joint 17th World Congress ofInternational Fuzzy Systems Association and 9th International Conference on Soft Computing and Intelligent Systems (IFSA- SCIS),Otsu, Japan, 2017, pp. 1-6, doi: 10.1109/IFSA- SCIS.2017.8023239.

[10] Bi, Z., Yu, L., Gao, H. et al. Improved VGG model-based efficient traffic sign recognition for safe driving in 5G scenarios. Int. J. Mach. Learn. & Cyber. (2020).

[11] Chuanwei Zhang et al., Study on Traffic Sign Recognition by Optimized Lenet-5 Algorithm, International Journal of Pattern Recognition and ArtificialIntelligence, doi:0.1142/S0218001420550034

[12] Han, C., Gao, G. & Zhang, Y. Real-time small traffic sign detection with revised faster-RCNN. Multimed Tools Appl78,1326313278(2019).

[13] H. S. Lee and K. Kim, "Simultaneous Traffic Sign Detection and Boundary Estimation Using Convolutional Neural Network," in IEEETransactions on Intelligent Transportation Systems, vol. 19, no. 5, pp.1652-1663, May 2018, doi: 10.1109/TITS.2018.2801560.

[14] R. Qian, Y. Yue, F. Coenen and B. Zhang, "Traffic sign recognitionwith convolutional neural network based on max pooling positions,"2016 12th International Conference on Natural Computation, Fuzzy Systems nd Knowledge Discovery (ICNC-FSKD), Changsha, China, 2016, pp. 578- 582, doi: 10.1109/FSKD.2016.7603237.

[15] A. Pon, O. Adrienko, A. Harakeh and S. L. Waslander, "A Hierarchical Deep Architecture and Mini-batch Selection Method for Joint TrafficSign and Light Detection," 2018 15th Conference on Computer and Robot Vision (CRV), Toronto, ON, Canada, 2018, pp. 102- 109,doi:10.1109/CRV.2018.00024.

[16] Saha S., Islam M.S., Khaled M.A.B., Tairin S. (2019) An Efficient Traffic Sign Recognition Approach Using a Novel Deep Neural Network Selection Architecture. In:Abraham A., Dutta P., Mandal J., Bhattacharya A., Dutta S. (eds) Emerging Technologies in Data Mining and Information Security. Advances in Intelligent Systems and Computing, vol 814. Springer, Singapore. 5_74

[17] A. Welzel, A. Auerswald and G. Wanielik, & quot; Accurate camera- based traffic sign localization," 17th International IEEE Conference on Intelligent Transportation Systems (ITSC), Qingdao, China, 2014, pp. 445-450, doi: 10.1109/ITSC.2014.6957730.

[18] M. Karaduman and H. Eren, & quot;Deep learning based traffic direction sign detection and determining driving style," 2017 International Conference on Computer Science and Engineering (UBMK), Antalya, Turkey, 2017, pp. 1046-1050, doi:10.1109/UBMK.2017.8093453.

[19] E. Winarno, W. Hadikurniawati and R. N. Rosso, & quot; Location based service for presence system using haversine method," 2017 International Conference on Innovative and Creative Information Technology (ICITech), Salatiga, Indonesia, 2017, pp. 1-4, doi:10.1109/INNOCIT.2017.8319153.

[20] Pal R, Ghosh A, Kumar R, et al. Public health crisis of road traffic accidents in India:Risk factor

assessment and recommendations on prevention on the behalf of the Academy of Family Physicians of India. J Family Med Prim Care. 2019;8(3):775-783. doi:10.4103/jfmpc.jfmpc_214_18