Inter-frame Arithmetic Device for Object Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In high-speed vision systems, the increasing number of pixels and higher resolution lead to a larger circuit scale and hindered sensing performance due to the need for arithmetic units corresponding to each pixel, resulting in increased system size and power consumption.
Innovation Solution
An arithmetic device with a frame memory and an inter-frame arithmetic processing unit that performs column-parallel arithmetic on pixel data between frames, updating past frame data based on the results, using OR and AND circuits for extension and extraction processes, reducing the hardware scale required for tracking objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If arithmetic units are provided one-to-one with each pixel to achieve accurate tracking, then tracking precision is improved, but circuit scale increases and sensing performance deteriorates due to increased system size
Solution Approach 1:
The patent divides the arithmetic processing into row-unit segments, where each row unit processes pixel data independently but uses shared arithmetic resources. This segmentation allows accurate row-by-row tracking while avoiding the need for full pixel-count arithmetic units, thus reducing circuit scale while maintaining tracking precision.
Solution Approach 2:
The patent merges multiple arithmetic units into a single shared arithmetic processing unit that operates on row-unit data. Instead of having separate arithmetic units for each pixel, the system combines them into one unit that processes data column-parallel across multiple pixels, reducing the total number of arithmetic units while maintaining processing capability.
2Measurement precision
If arithmetic units are provided one-to-one with each pixel to perform tracking arithmetic, then tracking function is improved, but power consumption increases due to larger system scale
Solution Approach 1:
The patent merges multiple pixel-level arithmetic units into a single shared arithmetic processing unit that handles row-unit data. This consolidation reduces the total number of active circuits and their associated power consumption while maintaining the tracking function through column-parallel processing of multiple pixels simultaneously.
Solution Approach 2:
The patent transitions from processing pixels sequentially (one-to-one correspondence) to processing pixels in parallel across columns (column-parallel). This dimensional change in processing architecture allows the system to maintain tracking accuracy while using fewer arithmetic units, thereby reducing power consumption.
3Measurement precision
If higher resolution and frame rates are implemented to improve sensing performance, then sensing accuracy is improved, but interface transfer band load increases and system becomes enlarged
Solution Approach 1:
The patent extracts the arithmetic processing function from the pixel-level and consolidates it into a shared row-unit processing architecture. This extraction allows the system to handle high-resolution and high-frame-rate data without proportionally increasing the number of arithmetic units, thus avoiding system enlargement while maintaining sensing performance.
Solution Approach 2:
The patent introduces column-parallel processing as an additional dimension to the traditional row-by-row processing. This allows the system to process multiple pixels simultaneously across columns, effectively handling higher resolution and frame rate data without linearly increasing system complexity or size.
Data Source
AI summary
The present disclosure relates to an arithmetic device that reduces a scale of an arithmetic processing unit which performs an arithmetic process between frames in a sensor. A frame memory stores pixel data of a frame that transitions in time sequence. An inter-frame arithmetic processing unit implements a predetermined arithmetic by column parallel in a row unit on the pixel data of a current frame and the pixel data of a past frame stored in the frame memory and updates the pixel data of the past frame stored in the frame memory on the basis of a result of the predetermined arithmetic.