SPAD-Based Camera Delivers High Dynamic Range Video

Single-photon avalanche diode (SPAD) sensors are photodetectors that generate a pulse when they detect a photon. SPADs are often used with LiDAR and time-of-flight (TOF) imaging systems. They tend to be very sensitive, allowing for low-light operation. The event-based nature of the diodes makes the output amenable to artificial-intelligence (AI) models connected to spiking neural networks (SNNs) that support neuromorphic computing, such as technology available from BrainChip.

I talked with Dr. Sebastian Bauer, CEO at Ubicept, about the company’s flexible light acquisition and representation engine (FLARE) smart camera that combines SPAD sensors with a GPU (Fig. 1). It delivers a video stream that exceeds the quality of conventional high speed, low light imaging systems. The GPU, which can also provide real-time AI analysis of the video stream, can be programmed to handle tasks like high-speed analysis, gated imaging, 3D perception, and even indirect sensing.

This type of camera is useful for many applications including automotive advanced driver-assistance systems (ADAS), in which the camera needs to operate in low-light environments (Fig. 2). Conventional cameras tend to provide lower-quality images because of tradeoffs and the frame-by-frame implementation.

SPAD systems have the advantage of providing an extremely high dynamic range. This is key in ADAS applications, where glare from lights and the sun can easily block out details when viewed from a video captured by a conventional, frame-based CMOS sensor.

The SPAD runs as an event-based sensor and the GPU can be used to generate a high-quality, frame-based data stream that would work with conventional AI models. It could also run an SNN, which tends to be more power-efficient.

The demonstration camera (Fig. 3) is large compared to what’s needed for most automotive applications. It’s designed for prototyping and evaluation, but a production system is considerably more compact.