Video surveillance systems have become important elements in aiding crime recognition. However, most rely on human observation for their operation and effectiveness, rendering them inefficient due to human limitations in monitoring. This work describes the development of a prototype that assists in the detection and recognition of weapons and intimidation positions in open spaces, using image processing and neural networks. The system captures images through videos received by the Jetson Nano development board, which are processed for detection and recognition using the Yolo Convolutional Neural Network (CNN) version 7. After the detection stage, an alert signal is sent to a mobile device, indicating whether an object has been recognized or whether an intimidation position associated with the carrying of firearms or sharp objects has been detected. The alert includes the image with the detected objects and the detection and transmission times, allowing video surveillance personnel to distinguish between risky positions and objects within the system’s field of view. Our results show that the accuracy in detecting handguns reached 91.13%, and that bladed weapons were 60.4%, under good lighting conditions. The average processing time per frame was 0.5 s, with an alert sending time of 0.3 s.