Object Detection Validation via Location Projection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object detection and classification methods in autonomous vehicles are computationally intensive and inefficient, as they continuously re-detect and re-classify objects that have already been correctly identified, leading to redundant processing and increased latency.
Innovation Solution
The method involves projecting and estimating future object locations based on current and past data, reducing the need for continuous detection and classification by performing a validation process, which can be executed as a background task or using cloud computing, thereby minimizing redundant processing and latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If continuous detection and classification is performed on all objects in the scene, then object detection accuracy is maintained, but computational complexity and processing time increase significantly
Solution Approach 1:
The system performs preliminary classification of objects into stationary and moving categories based on initial detection data. This preliminary action allows the system to establish different processing strategies for different object types before continuous tracking begins, reducing the need for repeated full classification operations.
Solution Approach 2:
The detection process is segmented into different stages: initial full scene classification, followed by selective tracking based on object motion characteristics. Stationary objects are handled differently from moving objects, with the system segmenting the processing workload to avoid redundant computations on objects that don't require continuous monitoring.
2Reliability
If continuous classification is performed on all detected objects, then classification accuracy is maintained, but processing latency increases
Solution Approach 1:
Instead of continuous classification, the system implements periodic validation where projected object locations are compared against actual detected locations at subsequent time steps. Classification is performed initially and then validated periodically, reducing the frequency of full classification operations while maintaining accuracy for objects that remain in expected positions.
Solution Approach 2:
The system extracts and separates the validation function from the full classification process. By taking out the location verification task and handling it independently through projection comparison, the system avoids the need to perform complete classification operations continuously, thereby reducing processing latency.
3Reliability
If redundant detection and classification operations are performed, then object tracking reliability is improved, but computational efficiency decreases
Solution Approach 1:
The system uses feedback from projected location comparisons to determine whether full re-detection and re-classification are necessary. When projected locations match actual detected locations, the system feedback-indicates that objects are stable and can continue with tracking only, avoiding redundant operations. When mismatches occur, full detection and classification are triggered to ensure reliability.
Solution Approach 2:
The processing intensity is made dynamic based on object behavior. Objects that maintain expected trajectories undergo lighter processing (tracking only), while objects that deviate from projections or are newly detected undergo full detection and classification. This dynamic adjustment of processing depth optimizes the balance between reliability and computational efficiency.
Data Source
AI summary
A system and method for object detection and classification includes receiving by a controller a first data of a scene, wherein the data reflects the scene at a first time, and performing a first classification of at least one object within the scene based on the data. The method includes determining a projected location of the at least one object corresponding to an estimated location at a second time. The method includes receiving a second data of the scene reflecting the scene at the second time. The method includes determining whether the projected location of the object corresponds to the location of the object as reflected by the second data. The method includes determining whether performing a second classification of the object is necessary based on the determination of whether the projected location corresponds to the location of the object reflected by the second data.