People detection and tracking

Object detection is a computer vision technique that identifies and locates objects within images or videos. It not only classifies objects into predefined categories—such as people, cars, or animals—but also draws bounding boxes around them to indicate their positions. Unlike simple image classification, which only determines the presence of an object, object detection provides both the “what” and the “where.” When extended to video sequences, object detection is often combined with tracking, a process that follows the movement of detected objects across multiple frames. Tracking assigns consistent identities to objects over time, allowing systems to monitor behavior, count objects, or predict motion. This combination is crucial in applications like autonomous vehicles, surveillance, and sports analytics, where understanding object dynamics is essential.

At NeuralSpike we developed a people detection system

which allows us to detect and track people on devices with low computational capabilities. Our system can be deployed on edge devices like System On Module (SOM) or even Microcontroller Units (MCU). We developed custom models based on YOLO or YOLOX architectures with low computational requirements and low memory footprint.
To develop such efficient models, we used our ml-toolbox, that allowed us to tune the architecture, with the architecture search technique. We also quantized the model to take advantage of Neural Processor Unit (NPUs) available on many embedded hardware. Our optimization process is flexible and we were able to tune our model to the hardware that it was deployed on. This way we were able to maximize the potential of the hardware platform in terms of speed, memory usage, and energy consumption.


Request your AI consultation…