SecurityAI

an AI image identification project that is a security system it can identify dangerous weapons and suspicious people, and then alerting users of danger

The AI detecting the position of the body and identifying the knife

Running this project

1. Make sure your Jetson Nano is connected to VSCode and you have camera connected to the nano
2. Make sure you have installed jetson-inference and python3
3. Download the labels.txt, securityAI-final.py, and the knife-rifledetect.onnx files
4. drag these three downoaded files into a new folder outside of the jetson-inference folder
5. navigate into the new folder containing the files in the terminal using cd <folder name>
6. run the securityAI-final.py using python3 securityAI-final.py
7. to view the output of the live camera, go to http://<nano-ip>:8554

The Algorithm

The explanation of the code is in the code below

one special thing is that the suspicion of a person is detected by calculating the ratio of the shoulders and wrists

you can modify this method if you want by using the cordinates from poseNet

#!/usr/bin/python3

from jetson_inference import detectNet
from jetson_inference import poseNet

from jetson_utils import videoSource, videoOutput

#importing needed things

net = detectNet(
    model="knife-rifledetect.onnx",
    labels="labels.txt",
    input_blob="input_0",
    output_cvg="scores",
    output_bbox="boxes",
    threshold=0.5
)

poseNet = poseNet("resnet18-body", threshold=0.15)
camera = videoSource("v4l2:///dev/video0")
display = videoOutput("webrtc://@:8554/output") # 'my_video.mp4' for file

#initializing AI and camera

while True:
    
    img = camera.Capture()
    
    if img is None: # capture timeout
        continue
    #camera feed
    
    detections = net.Detect(img)
    #detectNet detecting each frame
        
    for x in range(0,len(detections)):
        if(detections[x].ClassID == 1):
            print("Knife detected!")
        elif(detections[x].ClassID == 2):
            print("Rifle detected!")
    #using the feedback from detectNet to print what dangers were detected
    
    poses = poseNet.Process(img)
    #poseNet detecting pose points
    
    for pose in poses:
        # find the keypoint index from the list of detected keypoints
        # you can find these keypoint names in the model's JSON file, 
        # or with net.GetKeypointName() / net.GetNumKeypoints()
        left_wrist_idx = pose.FindKeypoint('left_wrist')
        left_shoulder_idx = pose.FindKeypoint('left_shoulder')
        right_wrist_idx = pose.FindKeypoint('right_wrist')
        right_shoulder_idx = pose.FindKeypoint('right_shoulder')

        # if the keypoint index is < 0, it means it wasn't found in the image
        if left_wrist_idx < 0 or left_shoulder_idx < 0 or right_wrist_idx < 0 or right_shoulder_idx < 0:
            continue
        
        left_wrist = pose.Keypoints[left_wrist_idx]
        left_shoulder = pose.Keypoints[left_shoulder_idx]
        right_wrist = pose.Keypoints[right_wrist_idx]
        right_shoulder = pose.Keypoints[right_shoulder_idx]
        #cordinates of each key position
        
        distance_wrist = (right_wrist.x - left_wrist.x)*(right_wrist.x - left_wrist.x)+(right_wrist.y - left_wrist.y)*(right_wrist.y - left_wrist.y)
        distance_shoulder = (right_shoulder.x - left_shoulder.x)*(right_shoulder.x - left_shoulder.x)+(right_shoulder.y - left_shoulder.y)*(right_shoulder.y - left_shoulder.y)
        #calculating the distance between shoulders and between wrists
        
        if(distance_wrist <= distance_shoulder*0.6):
            print("Suspicious person detected")
        #calculating the ratio of distances, if lower than the ratio, person is counted as suspicious, the ratio is currently 0.6
    
    display.Render(img)
    display.SetStatus("Object Detection | Network {:.0f} FPS".format(net.GetNetworkFPS()))