Pre-Processing Circuit for Event-Based Vision Sensors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing event-based vision sensors face challenges in reducing the computation amount of processing circuits, which leads to increased power consumption and complexity in processing image data.
Innovation Solution
A sensor device and semiconductor device are designed with a pre-processing circuit that performs weighted addition and determination processes on event signals from multiple pixel circuits, generating feature value signals to reduce the data rate and computation load for the processing circuit, including a neural network.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If event signals from multiple pixel circuits are directly processed by the neural network processing circuit, then the processing accuracy is maintained, but the computation amount and power consumption increase
Solution Approach 1:
The pre-processing circuit performs preliminary processing on event signals before they reach the neural network. Specifically, it generates feature value signals by comparing event signals from multiple pixel circuits and performs weighted addition on these feature values. This preliminary action reduces the computation burden on the neural network while preserving essential information for accurate processing.
Solution Approach 2:
The pre-processing circuit extracts essential features from raw event signals by generating feature value signals that represent the presence or absence of events at different positions. This extraction process separates the essential information needed for accurate processing from the raw data, reducing the computation amount while maintaining processing accuracy.
2Loss of information
If event signals are directly input to the neural network without pre-processing, then complete information is preserved, but the computation amount increases
Solution Approach 1:
The pre-processing circuit performs preliminary processing by generating feature value signals through comparison and weighted addition operations. This preliminary action condenses the information from multiple event signals into essential feature values, reducing the computation amount required by the neural network while preserving the necessary information for accurate processing.
Solution Approach 2:
The pre-processing circuit transforms the parameters of the input data by converting raw event signals into feature value signals with different characteristics. The weighted addition process changes the parameter representation from individual pixel events to aggregated feature values, reducing data dimensionality while maintaining information completeness.
3Measurement precision
If all event signals are processed by the neural network, then processing accuracy is maintained, but the processing time increases
Solution Approach 1:
The pre-processing circuit performs preliminary processing by generating feature value signals before neural network processing. This preliminary action reduces the data volume that the neural network must process, thereby reducing processing time while maintaining the accuracy of the final results through preserved essential features.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The solution effectively reduces the computation amount and power consumption of the processing circuit, enhancing processing efficiency and accuracy by preprocessing event signals before they reach the neural network.
Implementation Method 1
Each of the multiple pixel circuits includes a light receiver (photodiode) and is configured to generate an event signal corresponding to presence or absence of an event in accordance with a light reception result of the light receiver
Data Source
AI summary
A sensor device includes: multiple pixel circuits each of which includes a light receiver and generates an event signal corresponding to presence or absence of an event in accordance with a light reception result of the light receiver; and a pre-processing circuit that generates processing information on the basis of the multiple event signals, and supplies the processing information to a processing circuit including a neural network. The pre-processing circuit includes: a first weighted addition circuit that performs a weighted addition process on the basis of two or more of the event signals generated by respective two or more pixel circuits among the multiple pixel circuits; and a first determination circuit that generates a first feature value signal on the basis of a result of the weighted addition process in the first weighted addition circuit. The pre-processing circuit generates processing information based on the first feature value signal.