violence and crime detection(kidnapping, stealing) in public and educational milieu
- 0 Collaborators
This project entails in taking in consideration all the criminal and violent acts which occur within our society. This project came to life due to the rampant misbehaviors of young scholars in the school milieu such as aggressive attacks towards teachers and instructors , rape ...learn more
Project status: Under Development
Internet of Things, Artificial Intelligence
Groups
Internet of Things,
Student Developers for AI,
DeepLearning,
Cameroon Artificial Intelligence
Intel Technologies
DevCloud,
oneAPI,
Intel Integrated Graphics,
Intel FPGA,
Intel Python,
Intel vTune,
Movidius NCS,
Intel CPU,
OpenVINO,
Intel Opt ML/DL Framework
Overview / Usage
The problem which gave rise to this solution is that of the rampant aggressive misbehaviors of youths within the school milieu where they had to kill their teachers, aggressing them violently with arms such as knifes, we also get the rape situation between students and instructors. And my mother being a teacher all that terrified me. Added to that the country was also facing an increase in the number or rate of people been kidnapped for fraudulent actions or are been killed. Moreover theft which has always been a major problem to our society with enterprises, schools, homes facing such a lot of such situations. This system works in such that when a violence or criminal act is recognized within our society, school, enterprises or homes the authorities are actually informed about the criminal act accompanied with the location, an alert will be send to the security forces with an associated message and call forwarded to their lead . The crimes or violent action will be segmented into 3 categories, the first degree crimes will be send to the lower rank security forces while second and third degree will move to higher ranked security forces.
Methodology / Approach
In my proposed solution I make use of a deep neural network trained from scratch where I use a set of convolutional 3D layers with different max pooling layers for mean subtraction and scalability where I get a SoftMax of whether violence is recognized or not. Later on the model is been passed through several pruning methods and quantized per layer in order to archive an optimize model to work at the edge. when such is achieved, more so in my application code, I use a person detection model which takes in consideration that at least two person's in the captured set of frames in a stack size of 16 frames moves for inferencing in the violence model . This algorithm will have to work in match with the alarm device put in place using raspberry pi and an alert system which will give a precise location of where a particular violent action is going on in the society. Additional recognition such as theft and kidnapping will be added to the actual trained model SoftMax. Thereby creating kinds of segmentation techniques.
Technologies Used
openvino toolkit
DL workbench
vTune Amplifier
intel devcloud
intel oneapi
pytorch
tensorflow
python
opencv
person detection retail 0013
Docker