SecuritySpy

    Features — Smart motion detection powered by AI

SecuritySpy uses AI in the form of deep neural networks to provide extremely accurate triggering of recordings and notifications. Our bespoke neural networks have been meticulously trained on hundreds of thousands of real CCTV images, and perform the following functions:

1. An initial lightweight neural network ingests incoming video frames, predicting which ones contain real motion, along with the location of that motion within the frame. The accuracy of this network is over 95%, even in challenging conditions.

2. The image is then cropped to the area of motion, and sent to an image classifier, which is a larger neural network that predicts whether a human, vehicle or animal is present. The accuracy of this network for human and vehicle detection is over 97%.

With this advanced image processing, false triggers such as swaying plans, snow, rain and shadows are virtually eliminated.

All processing is done locally on your Mac, which protects your privacy, and allows for high performance and low latency.

Performance is very high on Apple Silicon Macs, due to the Neural Engine built into these chips. However, even Intel-based Macs can take advantage of these advanced networks, when used with moderate numbers of cameras.

Here's how it works:

		Incoming video frames are processed by a neural network that detects whether real motion motion is present, and where that motion is located. This is a bespoke network, designed and trained by us from the ground up, that is both highly accurate and extremely fast.
		When motion is detected, the image is cropped to just the area of motion, and scaled to a size that can be accepted by the classifier.
		The cropped image is passed to the classifier, which detects the presence of humans, vehicles and animals. This is a custom-trained version of ResNet-50, which is a powerful and accurate Convolution Neural Network.
		The AI outputs prediction values that indicate how likely it thinks there is a human, vehicle or animal in the image. You decide how this information is used to trigger various functions of the software: for example, you can set SecuritySpy to record all motion, but to generate notifications only when a human is detected.