uiux
Regular
US20230007162 - EVENT-BASED COMPUTATIONAL PIXEL IMAGERS
Applicants
Anduril Industries, Inc.
A computational pixel imaging device that includes an array of pixel integrated circuits for event-based detection and imaging. Each pixel may include a digital counter that accumulates a digital number, which indicates whether a change is detected by the pixel. The counter may count in one direction for a portion of an exposure and count in an opposite direction for another portion of the exposure. The imaging device may be configured to collect and transmit key frames at a lower rate, and collect and transmit delta or event frames at a higher rate. The key frames may include a full image of a scene, captured by the pixel array. The delta frames may include sparse data, captured by pixels that have detected meaningful changes in received light intensity. High speed, low transmission bandwidth motion image video can be reconstructed using the key frames and the delta frames.
From GPR:
The patent US20230007162 - Event-Based Computational Pixel Imagers describes an imaging system that integrates in-pixel signal processing, event-based detection, and multi-threaded computation. Unlike traditional CMOS or CCD image sensors that capture full frames at fixed intervals, this system detects and processes changes in light intensity at the pixel level. Each pixel contains a counter that increments and decrements based on detected intensity changes, allowing it to identify significant events while filtering out noise. The system employs a dual-frame acquisition strategy, capturing key frames at a lower frequency and delta frames at a higher frequency to efficiently track motion. Additionally, it incorporates infinite dynamic range counters, which utilize most significant bit (MSB) readout techniques to extend dynamic range without increasing counter bit-depth. This reduces data transmission requirements and enhances computational efficiency by prioritizing meaningful image data.
The architecture supports distributed in-pixel computation, including local filtering, thresholding, and adaptive exposure adjustments. Pixels can share data through orthogonal transfer mechanisms, enabling on-chip signal processing functions such as edge detection, temporal filtering, and convolution. The design also incorporates multi-threaded execution within the pixel array, allowing different processing tasks to run concurrently. The system's event-driven nature reduces power consumption and bandwidth usage by transmitting only data from pixels that detect significant intensity changes. These features align with neuromorphic imaging principles, as the system mimics biological vision by emphasizing changes in a scene rather than static information. However, the patent does not explicitly reference neuromorphic computing or spiking neural networks, instead focusing on the efficiency of event-based imaging and in-pixel digital processing.
posted an early-access link to an article called “Artificial Intelligence for Satellite Communication: A Survey”, co-authored by 23 SATCOM experts, the majority of whom - like herself - are affiliated with Uni Luxembourg’s Interdisciplinary Centre for Security, Reliability, and Trust (SnT).
.
Have a great 18.2.2025 
just the same as I love Arpan Pal ❤ 
