2024 BrainChip Discussion, page-8272

Some snippets of info from within the patent.

Performing the property reconstruction based on the difference between image raw data at one instance and a previous instance has the advantage that it may reduce the required processing time, since the calculations will be done only on the differential data rather than full data input.

Furthermore, the property reconstruction is done on the spikes representation of the data rather than the original data. This enables the use of spiking neural networks, which may benefit from the low power properties of the spiking neural networks and the advantageous use of neuromorphic processors.

The advantage of the described method is that event-based processing may be applied and that the raw data is processed if there is a change in the scene. This allows to lower the processing power and the temperature and power consumption of the system. If the sequence of images depicts a scene with a low rate of change, the number of processed images may drop down, and that the rate of processing new frames may drop down. Additionally, the lifetime of the components of the computing device implementing the method may be increased.

Also, it is possible to apply modern processor architectures based on neuromorphic computing, which may be optimized for event based processing and low power consumption.

The present application also relates to a computing device for image property reconstruction using a spiking neural network. The computing device comprises circuitry to execute any one of the previously described methods. Due to the event based processing of spiking neural networks, the lifetime of the components of the computing device may be increased. Also, it is possible to apply modern processor architectures based on neuromorphic computing, which may be optimized for event based processing and low power consumption.

The present application also relates to a lidar system comprising such a computing device and an image sensor. The image sensor is coupled to the computing device and the raw data of the image and/or the raw data of the preceding image comprises data provided by the image sensor.