Image Processing Device Object Detection Latency Reduction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image processing systems experience significant frame delay due to the need for object recognition processing to be performed outside the image sensor, leading to inefficiencies in object position detection and recognition processing.

Innovation Solution

An image processing device with a detection unit that identifies object positions within an input image, a generation unit that creates a recognition image at a predetermined resolution, and a recognition unit that performs object recognition on this image, allowing for faster processing within the image sensor.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Power

If object recognition processing is performed outside the image sensor in the application processor, then the recognition processing can be performed with sufficient computational resources, but significant frame delay occurs

Engineering Contradiction:
Improvecomputational resourcesVSAvoidframe delay
Core Design Contradiction:
PowerVSLoss of time

Solution Approach 1:

The patent divides the image processing workflow into two segments: (1) position detection using a lightweight model in the image sensor, and (2) detailed recognition using a full DNN model in the application processor. This segmentation allows the time-critical position detection to occur locally while the computationally intensive recognition can happen externally, resolving the contradiction between computational resources and processing speed.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system performs preliminary position detection using the lightweight model before transferring the image to the application processor. This preliminary action identifies and crops the region of interest, so that subsequent recognition processing in the application processor can focus only on the relevant area, reducing overall processing time while maintaining computational accuracy.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If the entire captured image is transferred to the application processor for recognition, then recognition accuracy can be maintained, but processing time increases significantly

Engineering Contradiction:
Improverecognition accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent extracts only the necessary region (object position and cropped image) from the entire captured image using the lightweight position detection model. By taking out only the relevant portion rather than processing the entire image, the system maintains recognition accuracy while dramatically reducing the data volume and processing time required for subsequent recognition operations.

Inventive Principle:
Principle #2Taking out (Extraction)

3Speed

If a lightweight model is used for position detection in the image sensor, then processing speed increases, but detection precision may be reduced

Engineering Contradiction:
Improveprocessing speedVSAvoiddetection precision
Core Design Contradiction:
SpeedVSMeasurement precision

Solution Approach 1:

The patent applies local quality by using different model complexities for different processing stages: a lightweight model for the time-critical position detection stage in the image sensor, and a more accurate full DNN model for the recognition stage in the application processor. Each stage uses the appropriate level of complexity for its specific requirement, optimizing both speed and precision locally.

Inventive Principle:
Principle #3Local quality

Data Source

PatentUS20240394905A1Image processing device and image processing method
Publication Date: 2024.11.28 SONY SEMICON SOLUTIONS CORP
  • US20240394905A1 patent drawing
  • US20240394905A1 patent drawing
  • US20240394905A1 patent drawing

AI summary

An image processing device according to an embodiment includes: a detection unit (201) that detects a position of an object, included in an input image, in the input image; a generation unit (200) that generates a recognition image having a predetermined resolution and including the object from the input image based on the position detected by the detection unit; and a recognition unit (204) that performs recognition processing of recognizing the object on the recognition image generated by the generation unit.