A 0.56V/0.8V Vision Sensor with Temporal Contrast Pixel and Column-Parallel Local Binary Pattern Extraction for Dynamic Depth Sensing Using Stereo Vision

Motion and depth information are crucial for features of position tracking and obstacle avoidance in diversified blooming applications like drone and AR/VR glasses. Therefore, the low-power low-latency vision sensor is in highly demand [1–5]. For motion detection (MD), there are asynchronous event-based [1] and the synchronous frame-based [2–5] approaches. In event-based approach, the redundant data traffic is reduced with the drawback of data collection time, special processing engine, and noise sensitivity. In frame-based approach, it stores and subtracts frame data with the need of in-pixel storage element and complex pixel circuit. To address these issues, this work presents a frame-based MD vision sensor with a new temporal contrast pixel (TCP), which realizes in-pixel temporal contrast calculation and event reporting with global-shuttering and frame-differencing pulse-width-modulation (PWM) operations using only 6 transistors and 1 capacitor (6T1C). Compared to the ping-pong structure in [5] using 10T2C, the proposed TCP can achieve more features including the consecutive event reporting in a 0.64x smaller pixel with a 2.7x higher fill factor (FF). Moreover, the LBP and ROI extraction are also implemented on-chip for disparity calculation and depth sensing in stereo vision system.