1st Level Project Presentation SEMINAR.pptx

1ST LEVEL PROJECT

1. Presented By: Vishwjeet Shinde : 122B5B299 Atharv Kulkarni : 123B2B317 Omkar Satardekar : 123B2B327 Renuka Patil : 123B2B324 Guided By Dr. Deepali N. Naik Real Time Surveillance Threat Detection System Pimpri Chinchwad Education Trust (PCET) Pimpri Chinchwad College of Engineering An ISO 9001:2015 Certified Institute, NBA Accredited, Accredited by NAAC with ‘A’ grade Department of Computer Engineering

2. PROBLEM STATEMENT Nowadays CCTV Surveillance is used to a greater extent but still it lacks the feature of automatic violence detection. Manual monitoring is not much of a feasible task and the time taken to respond to the situation is also crucial. So we need a real- time violence alert system to detect any violence and notify the concerned authorities with the required details in real- time. 1

3. MODULES HUMAN DETECTION VIOLENCE DETECTION IMAGE ENHANCEMENT ALERT SYSTEM 2

4. Detect humans Extract humans detected frames Recognize violent actions Enhance video frames Detect faces 6. Send alert to the concerned authority METHODOLOGY Faster RCNN Inception V2 COCO MobileNet V2 (Pretrained model) Blind Deconvolutional Algorithm YOLO V2 3 Python Libraries

5. OPERATING ENVIRONMENT PYTHON The language used. GOOGLE COLABORATORY Environment for running python and similiar Machine Learning and Deep Learning projects Able to use Google's GPU and TPU 4

6. ARCHITECTURAL DIAGRAM Violence Recognition Video frames Surveillance camera YES Human detection 5 Obtain the violence recognized frame Image Enhancement Discard frames NO YES NO Alert System YES Use of MobileNet Use of Blind Deconvolutional Algorithm , Yolo v2 Use of Pretrained models

7. USE CASE DIAGRAM Extract frames Violence detection Image Enhancement Alert System People involved in a violent activity Nearby Police station Real time Violence detection system Feed Real time video Detect humans inclu de 6 incl ude inc lude

8. RESULTS Faster RCNN Inception V2 COCO

9. RESULTS SSD Mobilenet V1 COCO

10. RESULTS 7 Three pre- trained models were compared and we obtained the following conclusions: Accuracy SSD Mobilenet V1 COCO < Faster RCNN Inception V2 COCO <= Faster RCNN Nas Speed SSD Mobilenet V1 COCO >= Faster RCNN Inception V2 COCO >> Faster RCNN Nas

11. METHODOLOGY Data Training Dataset Testing Dataset MobileNet v2 Model Testing Video Image frames

12. METHODOLOGY A dataset having 1000 videos each of violence category and non- violence category was chosen A model was trained using MobileNetV2 using the dataset Real- time video footage is given as input Output is obtained as image frames

13. MOBILENET V2 Convolutional neural network that is 53 layers deep Provides real- time classification capabilities under computing constraints in devices like smartphones. Utilizes an inverted residual structure where the input and output of the residual blocks are thin bottleneck layers. Uses lightweight convolutions to filter features in the expansion layer.

14. MOBILENET V2 ARCHITECTURE

15. RESULTS

16. ACTION PLAN AREA AND TOPIC RESEARCH Status : Completed LITERATURE REVIEW AND DATASET COLLECTION Status : Completed DESIGN & IMPLEMENTATION Status : Completed NOVELTY IMPLEMENTATION status : To be completed Nov Dec Apr Jan- Mar Implementation phase starts May FINAL TESTING, PAPER PUBLICATION Status : To be completed

17. REFERENCES Mi Young Lee, Ijaz Ul Haq, Seungmin Rho, Sung Wook Baik, and Samee Ullah Khan Cover the Violence: A Novel Deep- Learning- Based Approach Towards Violence- Detection in Movies, MDPI Article Received: 3 October 2019; Accepted: 7 November 2019; Published: 18 November 2019 M. - S. Kang, R. - H. Park and H. - M. Park, "Efficient Spatio- Temporal Modeling Methods for Real- Time Violence Recognition," in IEEE Access, vol. 9, pp. 76270- 76285, 2021, doi: 10.1109/ACCESS.2021.3083273, Date of Publication: 25 May 2021. Zhou P, Ding Q, Luo H, Hou X (2018) Violence detection in surveillance video using lowlevel features. PLoS ONE 13(10): e0203668. https://doi.org/10.1371/journal.pone.0203668, Published: October 3, 2018 https://towardsdatascience.com/review-mobilenetv2- light- weight- model- image- classification- 8febb490e61c

18. THANK YOU