Vehicle Object Detection with Adaptive Computation Omission
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection systems on vehicles face high computational loads, leading to potential failures in obtaining sensor signals at appropriate timing or executing appropriate processing, compromising vehicle safety.
Innovation Solution
An apparatus and method for object detection that omits part of the computational process based on vehicle speed, detected objects, distance, and situational factors, using a computation control unit to adjust the degree of omission to reduce computational load while ensuring safety.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep neural network is used for object detection, then detection accuracy is improved, but computational load increases enormously
Solution Approach 1:
The system dynamically adjusts the classification process based on detection results. When objects are detected in a sensor image, the system performs classification only on those detected objects rather than processing the entire image through a full deep neural network, thereby reducing computational load while maintaining detection accuracy for relevant objects
Solution Approach 2:
The system extracts only the detected objects from the sensor image for classification processing. Instead of applying the computationally intensive deep neural network to the entire image, it isolates and processes only the regions containing detected objects, significantly reducing the computational burden while preserving accurate classification for important targets
2Productivity
If sensor analysis information is used to adjust sensor rate, then processing efficiency is improved, but safety may be compromised when computation is omitted
Solution Approach 1:
The system uses detection results as feedback to control the classification process. When objects are detected, classification is performed on those objects; when no objects are detected, classification computation is omitted. This feedback mechanism ensures that safety-critical processing is maintained while avoiding unnecessary computation in safe conditions
Solution Approach 2:
The system performs classification only partially - specifically on detected objects rather than on all sensor data. This partial action approach maintains sufficient processing for safety-critical detected objects while omitting redundant classification computation for regions without objects, balancing efficiency and safety
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An apparatus for object detection includes an object detection unit (31) configured to detect an object around a vehicle (10) by inputting a sensor signal into a classifier. The sensor signal is obtained by at least one sensor (2, 3) for sensing the situation around the vehicle (10), and the sensor (2, 3) is mounted on the vehicle (10). The apparatus further includes a computation control unit (32) configured to set the degree of omission of computation so that at least part of computation on the sensor signal by the classifier will be omitted when the speed of the vehicle (10) is not greater than a predetermined speed threshold and the situation around the vehicle (10) does not correspond to any of predetermined exceptional situations.