www.turkjphysiotherrehabil.org 3528
A NEURAL NETWORK BASED APPROACH USED WITH DOWN-SAMPLING PROCESS TO COUNT NUMBER OF IMAGES
Dr. B. Dhanalaxmi1, M. Alekya2, V. Sai Nikhil3, G. Vamshi Krishna4
1Associate Professor, Department of Information Technology, Institute of Aeronautical Engineering, Dundigal, Hyderabad,Telangana, INDIA, [email protected]
2,3,4
Department of Information Technology,
Institute of Aeronautical Engineering, Dundigal, Hyderabad, Telangana, INDIA, [email protected]
ABSTRACT
Crowd counting is an dominant research topic in the sector of computer observation the multi-column convolution neural network. MCNN has been used in this sector and accomplish aggressive accomplishment.However when the crowd distribution is uneven the perfection of crowd counting based on the MCNN still needs to be improved in order to adapt to uneven crowd distributions.Crowd global density feature is taken into account in this system.Theglobal density aspects are draw out and added to the MCNN through the cascaded learning method because some comprehensive features in the course of the down- sampling process will be off track in the MCNN and it will affect the accuracy of the density map an improved MCNN structure is proposed in this system.The max pooling is replaced by max-ave pooling to keep more comprehensive features and the de-convolutional layers are added to restore the lost details in the down-sampling process the experimental results in the ucfcc50 data-set and the shanghai tech data-set show that the proposed method has higher accuracy and stability.
Keywords: - Global density feature, Deep learning, Convolutional neural network, Deconvolutional neural network, Down-sampling, Max-pooling, Max-Ave pooling, Crowd counting.
I. INTRODUCTION:
Crowd counting is used to compute the total swarm in a photo or video frame there are categorized into three types they are the direct counting method is based on target recognition.Whereas the indirect method is based on characteristic regression and deep learning the approach in the relevant researches based on target detection[8].This system initiate to use convolutional neural network to extract the characteristic area of the head- like contour and build to classify the characteristic area this initiated method to use physique contour of the body to achieve crowd detection and crowd density estimation all of these procedure are suitable for the scenes with a low-density crowd but the detection accuracy will decrease in the case of the high-density crowd in the relevant investigations. Based on characteristic regression the regression correlations between the image attributes and number of people for the crowd count proposed to use low-level characteristics and Bayesian regression to improve regression model robustness and flexibility[11].
Here utilize statistics from numerous sources to evaluate a single photo number of individuals and this study established a ucf cc 50 data set recent developments in profound learning and large numbers have progressively brought forward a multitude of counting systems based on profound[4] learning a cross-scene crowd counting approach has been suggested two learning targets the density map and the world number trained as a substitute the CNN single-column method is implemented however it was not appropriate in each branch network to alter the scale of the crowd that advocated using the mind with three branch networks to count the diverse scope and thus enhanced MCNN could adapt to alter the size of the population they introduced a new crowd counting data set which combines the features of shallow and convolution neural networks[6] to improve spatial resolution they proposed a multitasking network that combined previously high level with the density approximation suggest switch-CNN to crowd counting in this network.
www.turkjphysiotherrehabil.org 3529 A convolutional neural network and a well-monitored approach to delivering extra monitoring signalsuggested utilizing the lstm structure to extract crowd area contextual information suggestedadding an attention module to pick an adaptive counting mode for varied picture placementssuggested using the MCNN for robust counting of crowds in this study a density map has beencreated to increase the robustness of the crowd counting approach by taking comprehensiveinformation and size variations into consideration[5]
II. EXISTING SYSTEM:
Crowd counting is used to compute the total swarm of people in a photo or video frame.proposed to use haar wavelet transform to extract the feature area of the head-like contour and build the SVM classifier to classify the feature area[7].Thismethod used to shape contour of the body to achieve crowd detection and crowd density estimation all of these procedure are suitable for the scenes with a low-density crowd but the detection accuracy will decrease in the case of the high-density crowd in the relevant investigations based on feature regression the regression correlations between the image attributes and number of people for the crowd count.proposed to use low-level characteristics[10] and Bayesian regression to improve regression model robustness and flexibility.
III. RELATED WORK:
Physics-based deep learning,image recognition and direct counting and the statistical models that we have used to count the total number of images in photo or video frame.The maximum count to the people is estimated in a single image. We use the Convolutional neural network and the Multi covolutional neural network that we are used to estimate maximum count[9]. The neural network concept is used to count the total crowd by total count estimation.
IV. LITERATURE SURVEY:
Sheng-Fuu Lin, and his team proposed the Estimation of Number of People in Crowded Scenes Using Perspective Transformation[14]. In this paper,the developed system goes one step further to estimate the number of people in crowded scenes in a complex background by using a single image. Therefore, more valuable information than crowd density can be obtained. There are two major steps in this system: recognition of the head-like contour and estimation of crowd size. First, the Haar wavelet transform (HWT) is used to extract the featured area of the head-like contour, and then the support vector machine (SVM) is used to classify these featured area as the contour of a head or not. Next, the perspective transforming technique of computer vision is used to estimate crowd size more accurately[12]. Finally, a model world is constructed to test this proposed system and the system is also applied for real-world images.
XiaohangXu, and his team proposed Crowd Density Estimation of Scenic Spots Based on Multi feature Ensemble Learning [13]. In this system, the Estimating the crowd density of public territories, such as scenic spots, is of great importance for ensuring population safety and social stability. Due to problems in scenic spots such as illumination change, camera angle change, and pedestrian occlusion, current methods are unable to make accurate estimations. To deal with these problems, an ensemble learning (EL) method using support vector regression (SVR) is proposed in this study for crowd density estimation (CDE). The method first uses human head width as a reference to separate the foreground into multiple levels of blocks. Then it adopts the first-level SVR model to roughly predict the three features extracted from image blocks, including D-SIFT, ULBP, and GIST, and the prediction results are used as new features for the second-level SVR model for fine prediction. The prediction results of all image blocks are added for density estimation according to the crowd levels predefined for different scenes of scenic spots. Experimental results demonstrate that the proposed method can achieve a classification rate over 85% for multiple scenes of scenic spots, and it is an effective CDE method with strong adaptability[15].
Min Li, Zhaoxiang Zhang, and his team proposed the Number of People in Crowded Scenes by MID Based Foreground Segmentation and Head-shoulder Detection. In this system, proposes a novel method to address the problem of estimating the number of people in surveillance scenes with people gathering and waiting.The proposed method combines a MID (Mosaic Image Difference) based foreground segmentation algorithm and a HOG (Histograms of Oriented Gradients) based head-shoulder detection algorithm to provide an accurate estimation of people counts in the observed area[16]. In our framework, the MID-based foreground segmentation module provides active areas for the head-shoulder detection module to detect heads and count the number of people. Numerous experiments are conducted and convincing results demonstrate the effectiveness of method.
www.turkjphysiotherrehabil.org 3530 Tao Zhao and his team proposed Segmentation and Tracking of Multiple Humans in Crowded Environments. In this system segmentation and tracking of multiple humans in crowded situations is made difficult by inter object occlusion.In this system propose a model-based approach to interpret the image observations by multiple partially occluded human hypotheses in a Bayesian framework.This method define a joint image likelihood for multiple humans based on the appearance of the humans, the visibility of the body obtained by occlusion reasoning, and foreground/background separation[19]. The optimal solution is obtained by using an efficient sampling method, data-driven Markov chain Monte Carlo (DDMCMC), which uses image observations for proposal probabilities.
Knowledge of various aspects, including human shape, camera model, and image cues, are integrated in one theoretically sound framework. We present experimental results and quantitative evaluation, demonstrating that the resulting approach is effective for very challenging data.
Xiangjie Kong and his team proposed long-term traffic anomaly detection based on crowd sourced bus trajectory data. In this system as the development of crowd sourcing technique, acquiring amounts of data in urban cities becomes possible and reliable, which makes it possible to mine useful and significant information from data Traffic anomaly detection is to find the traffic patterns which are not expected and it can be used to explore traffic problems accurately and efficiently[20]. The team proposesLoTAD to explore anomalous regions with long-term poor traffic situations. Specifically, this method process crowd sourced bus data into TS-segments (Temporal and Spatial segments) to model the traffic condition. Later, explore anomalous TS-segments in each bus line by calculating their AI (Anomaly Index). Then, we combine anomalous TS-segments detected in different lines to mine anomalous regions. The information of anomalous regions provides suggestions for future traffic planning. Conduct experiments with real crowd sourced bus trajectory datasets of October in 2014 and March in 2015 in Hangzhou. We analyze the varieties of the results and explain how they are consistent with the real urban traffic planning or social events happened between the time intervals of the two datasets[21]. The team do a contrast experiment with the most ten congested roads in Hangzhou, which verifies the effectiveness of LoTAD.
V. PROPOSED METHODOLOGY:
In the proposed system were these algorithms have good performances in the crowd counting, but the performances of these methods were not effective when the crowd distribution is uneven .To solve the problem of inaccurate counting caused by uneven crowd distribution, the global density feature is extracted and used in this project.A convolutional neural network with global density feature by using multi-task network cascades (MNCs)is proposed. In order to generate a more comprehensive density map, the max-ave pooling layers are used to keep more features of the image. Meantime, the deconvolutional layers are added to the convolutional neural network in order to restore the lost details in down-sampling process. It will help to improve the accuracy of density map and further improve the accuracy of crowd counting
In this system our team uses convolutional neural networks to count number of people in an image or a video frame in neural networks we use the deep learning algorithm.It has,two layer techniques they are:
1 Convolutional layers 2 De-Convolutional layers
In convolutional layer itextracts the data of the image and sent for both density classification sub task and the crowd counting sub task.The density classification sub task as the global density feature it extracts the features of and the crowd counting sub task as the crowd density feature these two are combined and down to the de convolutional layer in the deconvolutional layer the image size is decreased and the images and formed. These images are formed dot like structured in the estimated density map.
www.turkjphysiotherrehabil.org 3531 Fig-1 System Arcitecture of convolutional neural network with global density feature
Fig-2 Count prediction graph
The estimation of crowd from the image the estimated count of crowd from 250 to 500 in our proposed system.The above map represents the density map were the estimated density map is shown.The original count can be detected but sometimes the count prediction may increase because some of the objects are get into the count.
We import the data from the data set. All the data present in the data set is generated were these data set describes the estimation of crowd in the given photo.
Modules used in our project:
K-means clustering algorithm:
www.turkjphysiotherrehabil.org 3532 KNN Algorithm:
K Nearest algorithm combines k closet points based on their extent and joins them in a cluster and these clusters are then evaluated. KNN are all supervised learning algorithms: decisions trees, neural networks, and their overall objective equal to predict new data accurately after training on existing data. We are given tuples in training (x1 ...xn, y). Classification mistake of data not encountered during the training phase is a testing mistake.
Artificial Neural Network:
Among them determiner approved ml methods is artificial neural networks as the neuronal detail of their name advocate they are brain stimulate layout that are programmed to imitate the approach that we humans appreciate countless neural networks have an insert and product flim as well as an invisible flim of units that alter the insert into integers that the product flim can recognize
The density classification sub-task is constructed by four convolutional layers; the number and size of convolutional kernels are represented by 16 × 9 × 9, 32 × 7 × 7, 16 × 7 × 7, 8 × 7 × 7. These four convolutional layers are used for global features extraction, and the extracted global features are used for both density classification and crowd counting. After that, the extracted global features are fed back to adaptive max pooling layer and fully connected layer. The adaptive max pooling layer is used to get fixed size features. The last fully connected layer has five neurons for estimating the density level.
Results and discussions:-
Fig-3 Code Implementation
www.turkjphysiotherrehabil.org 3533 Fig-4 Crowd Counting Estimation Map
The estimated crowd count is seen in the above pictures the data is loaded and compiled were the process has been done internally. As the objects are detected as the person were the sometimes the objects are detected in the density map of the crowd.
VI. CONCLUSION:
An improved convolutional neural network combined with a global density feature is proposed in this system. It differs from current crowd-counting methods. The proposed method is concerned with uneven crowd distribution. However, when the crowd distribution is uneven, the perfection of crowd counting based on the MCNN still needs to be improved in order to adapt to uneven crowd distributions. This system takes crowd global density into account.Furthermore, the max-ave pooling and deconvolutional layers are employed to produce a more comprehensive density map. The experimental results show that the proposed method performs well on various crowd datasets. Because of the dense crowd, some backgrounds will be misidentified as people. It will introduce noise into the estimated density map and have an impact on the counting results.
VII. FUTURE SCOPE OF STUDY:
It will bring about noise in the allusive solidity map and influence the compute results for the future work we will focus on shrink the noise in the approximate solidity map and improving the preciseness.
REFERENCES:
1 S.-F. Lin, J.-Y.Chen, and H.-X. Chao, „„Estimation of number of people in crowded scenes using perspective transformation,‟‟ IEEE Trans.Syst., Man, Cybern. A, Syst., Humans, vol. 31, no. 6, pp. 645–654, Nov. 2001.
2 P. Gardzinski, K. Kowalak, L. Kaminski, and S. Mackowiak, „„Crowd density estimation based on voxel model in multi-view surveillancesystems,‟‟
in Proc. Int. Conf. Syst., Signals Image Process. (IWSSIP), Sep. 2015, pp. 216–219.
3 M. Li, Z. Zhang, K. Huang, and T. Tan, „„Estimating the number of people in crowded scenes by MID based foreground segmentation and headshoulderdetection,‟‟ in Proc. 19th Int. Conf. Pattern Recognit., Dec. 2008, pp. 1–4.
4 T. Zhao, R. Nevatia, and B. Wu, „„Segmentation and tracking of multiple humans in crowded environments,‟‟ IEEE Trans. Pattern Anal. Mach.
Intell.,vol. 30, no. 7, pp. 1198–1211, Jul. 2008.
5 X. Kong, X. Song, F. Xia, H. Guo,J. Wang, and A. Tolba, „„LoTAD: Longterm traffic anomaly detection based on crowdsourced bus trajectory data,‟‟ World Wide Web, vol. 21, no. 3, pp. 825–847, May 2018.
6 B.Dhanalaxmi, “Machine learning and its emergence in the modern world and its contribution to artificial intelligence”, 2020 International Conference forEmerging Technology, INCET 2020, art. no. 9154058. doi:10.1109/INCET49848.2020.9154058
7 D.Nikhil, Dr.K.Srinivasa Reddy, B.Dhanalaxmi, “Image processing based cancer detection techniques using modern technology - A survey”, Proceedings of the 5th International Conference on Communication and Electronics Systems, ICCES 2020, 2020, pp. 1279–1282, 09137997
www.turkjphysiotherrehabil.org 3534
8 D.Nikhil, B.Dhanalaxmi, Dr.K.Srinivasa Reddy, “The evolution of cloud computing and its contribution with big data analytics”, Lecture Notes on Data Engineering and Communications Technologies, 46, pp. 332-341. doi: 10.1007/978-3-030-38040-3_38
9 B. Dhanalaxmi, G. Apparao Naidu, and K. Anuradha, “A Rule Based Prediction Method for Defect Detection in Software System,” Journal of Theoretical and Applied Information Technology, Vol. 95, Number 14, 31st July 2017, pp 3403-3412.
10 B. Dhanalaxmi, G. Apparao Naidu, and K. Anuradha, “A Survey on Design and Analysis of Robust IOT Architectute”, International Conference on Innovative Mechanisms for Industry Applications, 13th July 2017, pp 375-378, DOI:10.1109/ICIMIA.2017.7975639
11 B. Dhanalaxmi, G. Apparao Naidu, and K. Anuradha, “Adaptive PSO based Association Rule Mining Technique for Software Defect Classification using ANN”, International Conference on Information and Communication Technologies, Procedia Computer Science, Vol. 46,2015, pp 432-442 12 B. Dhanalaxmi, G. Apparao Naidu, and K. Anuradha, “Defect Classification using Relational Association Rule Mining based on Fuzzy Classifier
along with Modified Articial Bee Colony Algorithm,” Indian Journal of Applied Engineering Research, Vol. 12, Number 11,June 2017, pp 2879-2886 13 B. Dhanalaxmi, G. Apparao Naidu, and K. Anuradha, “A Fault Prediction Approach based on the Probabilistic Model for Improvising Software
Inspection,” Indian Journal of Science and Technology, Vol. 9, Issue 45, December 2016.
14 Dhanalaxmi, G. Apparao Naidu, and K. Anuradha, “A Review on Software Fault Detection and Prevention Mechanism in Software Development Activities,” Journal of Computer Engineering, Vol. 17, Issue 6, pp. 25 - 30, Nov – Dec. 2015.
15 B. Dhanalaxmi, G. Apparao Naidu, and K. Anuradha, “Practical Guidelines to Improve Defect Prediction Model – A Review”, International Journal of Engineering Science Invention, Vol. 5, Issue 9, pp. 57-61, September 2016.
16 B. Dhanalaxmi, Dr. G.AppaRao, Naidu and Dr.K. Anuradha, “A Survey on Software Inspection Improvisation Techniques through Probabilistic Fault Prediction Method”, Journal of Advanced Research in Dynamical and Control Systems, Vol. 10,Special Issue.7 june 2018 pp. 617-621
17 B. Dhanalaxmi, Dr.G.AppaRao, Naidu and Dr.K. Anuradha, “A Review on Different Defect Detection Models in Software Systems”, Journal of Advanced Research in Dynamical and Control Systems, Journal of Advanced Research in Dynamical and Control Systems, Vol. 10,Special Issue.7 june 2018 pp.241-243
18 B.Dhanalaxmi, IndurthiRavindra Kumar, Vishal and B Madhuravani, “An User-Friendly Recital Teller System”, International Journal of Pure and Applied Mathematics, Vol. 119 No. 16 May 2018, pp. 1801-1805
19 D. B. Sam, S. Surya, and R. V. Babu, „„Switching convolutional neuralnetwork for crowd counting,‟‟ in Proc. IEEE Conf. Comput. Vis.
PatternRecognition., Jul. 2017, pp. 4031–4039.
20 Z. Shi, L. Zhang, Y. Sun, and Y. Ye, „„Multiscale multitask deep NetVLADfor crowd counting,‟‟ IEEE Trans. Ind. Informat., vol. 14, no. 11,pp.
4953–4962, Nov. 2018.
A. B. Chan and N. Vasconcelos,„„Counting people with low-level features and Bayesian regression,‟‟ IEEE Trans. Image Process., vol. 21, no.
4, pp. 2160–2177, Apr. 2012.
21 H. Idrees, I. Saleem, C. Seibert, and S. Mubarak, „„Multi-source multi-scale counting in extremely dense crowd images,‟‟ in Proc. IEEE Conf.
Comput. Vis. Pattern Recognit., Jun.2013, pp. 2547–2554.
22 W. Ma, L. Huang, and C. Liu,„„Crowd density analysis using co-occurrence texture features,‟‟ in Proc. 5th Int. Conf. Comput. Sci. Converg.Inf.
Technol., Nov./Dec. 2010,pp. 170–175.
23 D. Kong, D. Gray, and H. Tao, „„A viewpoint invariant approach for crowd counting,‟‟ in Proc. 18th Int. Conf. Pattern Recognit., Aug. 2006, pp.
1187–1190.
24 X. Kong, F. Xia, Z. Ning, A. Rahim, Y. Cai, Z. Gao, and J. Ma, „„Mobility dataset generation for vehicular social networks based on floating cardata,‟‟ IEEE Trans. Veh.Technol., vol. 67, no. 5, pp. 3874–3886, May 2018.
