PLMR-3P-PEODET

Plumerai makes deep learning tiny and radically more efficient to enable inference at reduced compute needs on low cost and low-power consumption hardware. Plumerai focuses on full stack, and has offices in London, Amsterdam and Warsaw.

Plumerai has developed a complete software solution for camera-based people detection. This people detection AI model is capable of running at 22FPS on TI AM62x processors with one of four available Arm Cortex-A cores, which allows three Arm A53 core to be available for other application specific compute needs. Plumerai continuously improves the reliability and accuracy of models and current models are trained with over 30 million images, which provides very high accuracy under a wide variety of conditions and scenarios - indoor or outdoor, low light or bright sunlight, robust detection against other moving objects or difficult backgrounds, which make this people detection reliable.

Target applications include security cameras, video doorbells, smart home cameras, people counting, auto-lock for laptops, auto-framing for video conferencing, elderly care, and building management systems such as HVAC, lighting control and occupancy detection.

Functionality:

• Detects each person in view, even if partially occluded
• Tracks people across frames and assigns up to 20 unique IDs
• Supports indoor, outdoor, low light and difficult backgrounds such as moving objects
• Detects at more than 20m / 65ft distance
• Trained with >30 million images
• Extensively validated with data unit tests, warranting robust performance on diverse people and settings

Input:

• QVGA to 4K video stream
• RGB or YUV format
• Color, greyscale, NIR input
• Standard and wide-angle lenses
• Supports many image sensors

Output:

• Bounding box coordinates
• Unique ID per person