AU2021107321A4

AU2021107321A4 - Crime prediction and prevention using computer vision and deep learning model

Info

Publication number: AU2021107321A4
Application number: AU2021107321A
Authority: AU
Inventors: Sharvil Kishore Darne; I. Diana Jeba Jingle; Pavan P. Kashyap; Renisha P. S.; P. Mano Paul; P. Sam Paul; Shirley Josephine Mary R.; D. S. Shylu; Brijesh Sathian; S. Thilagamani
Original assignee: Thilagamani S Dr
Current assignee: Thilagamani S Dr
Priority date: 2021-08-25
Filing date: 2021-08-25
Publication date: 2021-12-16
Anticipated expiration: 2029-08-25

Abstract

Crime detection is one of the most useful applications for deep learning because it helps curb crime and increase people's security. In this disclosure, a system is developed for object detection using YOLO Darknet model for the detection of suspicious objects like weapons and knives in the hands of an individual. This ensures that false alarms can be prevented by only detecting a weapon or knife. The model will help to detect and predict the crimes. The proposed model is implemented using Convolutional Neural Network with YOLO V3 model and this has achieved an better accuracy than other existing algorithms. The model has also shown that how deep learning model can perform better than the machine learning model in the case of object detection. 9 Descriptive analysis of data Cleaning and transformation of data Modelling of data Performance estimation FIGURE 3 Model Accuracy (%) Decision tree 59.15 K-NN (K=10) 87.53 Traditional CNN 94.3 Faster RCNN 95.44 YOLO V2 96.76 YOLO V3 98.89 FIGURE 4

Description

Descriptive analysis of data

Cleaning and transformation of data

Modelling of data

Performance estimation

FIGURE 3

Model Accuracy (%) Decision tree 59.15

K-NN (K=10) 87.53

Traditional CNN 94.3 Faster RCNN 95.44

YOLO V2 96.76

YOLO V3 98.89

FIGURE 4

Crime prediction and prevention using computer vision and deep learning model

FIELD OF THE INVENTION

The system and the method for crime prediction and prevention are detailed in this disclosure which involves computer vision and deep learning-based approaches.

BACKGROUND OF THE INVENTION

Computer vision is a branch for training a computer in recognizing to grasp a virtual object or word and so gives a feel to comprehend the environment. It largely analyses surrounding data from a camera and is hence important for its applications. They can be used to recognise the face, recognise plaques for numbers, increase and blend realities, locate and identify objects. Research on mathematical strategies are currently being undertaken for the recovery and computer comprehension of 3D images. Obtaining 3D object visuals can help us with the detection of objects, pedestrian recognition, face-recognition, active look, personal. These are merely basic applications and can be studied further in any of the categories indicated above. The algorithms is introduced for undertaking quick prototyping research into computer vision such that a tool is much faster than anticipated in order to achieve computer vision outcomes. Human posture can also be identified with regard to facial detection / human recognition.

Machine Learning (ML) allows to learn from prior events and improve them automatically without having to be explicitly programmed. After evaluating the data, it is not always possible to discern an accurate pattern or information. The exact pattern and information is used to interpret the ML in such circumstances. ML advocates the idea that the machine may learn and resolve complicated mathematical issues and some specialised problems.

The machine is trained in supervised learning, that enable it to achieve correct conclusions. The machine has a series of data in unattended learning and has to come across certain patterns as well as connections between the data. Neural networks have been studied since the 1980s and are significant tools for supervised learning. The researcher also suggests that several elements are necessary to achieve an exit from the completeness of the non-deterministic polynomial (NP) and that architectural restrictions are not enough.

SUMMARY OF THE INVENTION

It is simple to forecast crimes before they occur, yet the concept needs much more than to realize. This study was designed to help researchers build and utilize advanced technologies in real life to execute crime prediction. This study set out a foundation for developing a system that is much better for police forces. The initial problem is making this system effective, followed, among other things. A notable feature of this invention is that we have implemented YOLO (You Only Look Once) model with Convolutional Neural Network for the object detection. The model has achieved better accuracy than the traditional CNN models.

This disclosure has a second objective: applying the deep learning method for detecting the crime, that can prevent us from further happening any crime.

The disclosures outlined above present the modes of operation that can be combined with DL to assist in identifying and detecting the crime.

This disclosure is also made to show how effective the deep learning model is in comparison to the other machine learning algorithms already on the market.

Detailed drawings and additional information on the invention can be found in the attachment. The scope of the invention is not limited by the above illustrations, which only show the most common aspects of the storey. Specificity and detail will be provided alongside the drawings to supplement the explanation.

BRIEF DESCRIPTION OF FIGURES

When the following thorough explanation is read concerning the accompanying drawings, where like characters indicate like parts throughout the drawings, these and other features, aspects, and advantages of the current disclosure will become clearer: Figure 1 illustrates a flow diagram of a method for crime prediction and prevention, Figure 2 illustrates a data flow diagram of a system in which collection of images based on different types of crimes is predicted by using deep learning models for the analysis of crime detection, Figure 3 illustrates a flow diagram for the functionality approach of the proposed model. This includes the data cleaning and the transformation before modeling the data. Finally the performance is estimated using the parameter i.e. accuracy. Figure 4 illustrates the comparison table of the models in which machine learning algorithms are compared with deep learning models on the basis of parameters i.e. accuracy. The table has also shown the comparison of different types of YOLO models. The YOLO V3 has achieved the accuracy of 9 8 .8 9 % which is better than the other states of the art algorithms.

Figure 5 illustrates the comparison graph based on the accuracy of the machine learning models and the deep learning models. The graph has also shown that the deep learning models performed better than the machine learning algorithm in the case of object detection and the object prediction.

Further, skilled artisans will appreciate that elements in the drawings are illustrated for simplicity and may not have been necessarily been drawn to scale. For example, the flow charts illustrate the method in terms of the most prominent steps involved to help to improve understanding of aspects of the present disclosure. Furthermore, in terms of the construction of the device, one or more components of the device may have been represented in the drawings by conventional symbols, and the drawings may show only those specific details that are pertinent to understanding the embodiments of the present disclosure so as not to obscure the drawings with details that will be readily apparent to those of ordinary skill in the art having benefit of the description herein.

DETAILED DESCRIPTION:

Embodiments of the present invention will be described below in detail with reference to the accompanying drawings.

The new features depend entirely on the computer and thus require human development interaction, but once developed, they function without interaction and release people for other tasks. The proposed method has been implemented in a variety of research using well-known or state-of-the-art designs. Some research; on the other hand, have presented their own customized algorithm and architecture, which are not based on well-known architectures. In this disclosure we have used the YOLO object detection model for detecting the crime scene. Previously deep learning model i.e. Convolutional Neural

Network (CNN) is used for the classification, but it cannot be considered a better model for the detection purpose. So, for performing the detection, a YOLO model is proposed with CNN.

Figure 1 illustrates a flow diagram of a method for crime prediction and prevention, wherein the method compromises of: Step (102) discloses about capturing an image of surrounding using a capturing module.

Step (104) discloses about processing captured image using a neural network having a plurality of filters for producing a first output image, wherein adding a first pooling layer to the first output image, wherein the neural network is a convolution neural network (CNN) having 16 filters of size 3X3 having a first input size is selected from but not limited to 416x416X3 to obtain the first output image of size is selected from but not limited to 416x416X16.

Step (106) discloses about reprocessing the first output image using the neural network to obtain a second output image, wherein adding a second pooling layer to the second output image, wherein the CNN includes a second input size is selected from but not limited to 208 x 208 x 16 to obtain the second output image of size is selected from but not limited to 208 x 208 x 32, wherein a size of the pooling layer is selected from but not limited to 2X2.

Step (108) discloses about predicting multiple bounding boxes and class probabilities of boxes of the first output image and the second output image using the neural network, wherein predicting the crime based on the predicted bounding boxes, wherein a training model of the CNN you only look once (YOLO) trains the captured photos and optimizes detection performance by identifying and eliminating false positives.

Referring to Figure 2, a data flow diagram of a system for the prediction of crime scene is implemented. In the first step data is collected and preprocessed. Then in the second phase, model is selected i.e. Convolutional Neural Network for performing the training and testing and finally fine tuning is performed by YOLO models. This YOLO model is finally used for detection and prediction of the crime images.

All the implementation is done through python programming language. The crime related image dataset is split into two parts i.e. training dataset and the testing dataset. The model is developed using the training dataset and the validation is done through the testing dataset. The model used is Convolutional neural networks which consist of different types of dropout layer, regularization method and the max pooling layers. Finally the model is tuned using the YOLO V3 model.

Figure 3 illustrates a flow diagram of a method for the functional architecture of the proposed model. The descriptive analysis of the data is done in the first step, followed by the cleaning and the transformation of the data. Finally the modeling of the data is implemented with the performance evaluation of the model.

Figure 4 illustrates the comparison table of these models based on the parameters such as accuracy. The models used for the comparison are decision tree, K-NN, CNN, and the YOLO V2 model. It can be observed that YOLO V3 has performed better than the other existing models. The model has achieved the accuracy of 9 8 . 8 9 %. So it can be easily said that YOLO V3 is better than the other states of the art algorithms.

Figure 5 illustrates the comparison graph of the proposed model and the existing classifiers. YOLO V3 has achieved the accuracy of 98.89%.

The drawings and the forgoing description give examples of embodiments. Those skilled in the art will appreciate that one or more of the described elements may well be combined into a single functional element. Alternatively, certain elements may be split into multiple functional elements. Elements from one embodiment may be added to another embodiment. For example, orders of processes described herein may be changed and are not limited to the manner described herein. Moreover, the actions of any flow diagram need not be implemented in the order shown; nor do all of the acts necessarily need to be performed. Also, those acts that are not dependent on other acts may be performed in parallel with the other acts. The scope of embodiments is by no means limited by these specific examples. Numerous variations, whether explicitly given in the specification or not, such as differences in structure, dimension, and use of material, are possible. The scope of embodiments is at least as broad as given by the following claims.

Benefits, other advantages, and solutions to problems have been described above with regard to specific embodiments. However, the benefits, advantages, solutions to problems, and any component(s) that may cause any benefit, advantage, or solution to occur or become more pronounced are not to be construed as a critical, required, or essential feature or component of any or all the claims.

Claims

WE CLAIM:

1. A method for crime prediction and prevention, wherein the method compromises of: capturing an image of surrounding using a capturing module; processing captured image using a neural network having a plurality of filters for producing a first output image, wherein adding a first pooling layer to the first output image; reprocessing the first output image using the neural network to obtain a second output image, wherein adding a second pooling layer to the second output image; and predicting multiple bounding boxes and class probabilities of boxes of the first output image and the second output image using the neural network, wherein predicting the crime based on the predicted bounding boxes.

2. The method as claimed in claim 1, wherein the neural network is a convolution neural network (CNN) having 16 filters of size 3X3 having a first input size is selected from but not limited to 416x416X3 to obtain the first output image of size is selected from but not limited to 416x416X16.

3. The method as claimed in claim 1, wherein the CNN includes a second input size is selected from but not limited to 208 x 208 x 16 to obtain the second output image of size is selected from but not limited to 208 x 208 x 32.

4. The method as claimed in claim 1, wherein a size of the pooling layer is selected from but not limited to 2X2.

5. the method as claimed in claim 1, wherein a training model of the CNN you only look once (YOLO) trains the captured photos and optimizes detection performance by identifying and eliminating false positives.

FIGURE 2 FIGURE 1

FIGURE 3

FIGURE 4

FIGURE 5