Neural Object Detection With Batched Property Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in efficiently selecting locations where objects are likely centered and determining their properties in real-time, leading to latency and reduced accuracy in driving decisions.
Innovation Solution
A neural network system that uses sensor data to determine object scores for potential object locations in a single forward pass, allowing for simultaneous selection of object locations and subsequent determination of object properties in batches, thereby reducing latency and improving computational efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the system processes each location sequentially to determine object scores and properties, then computational accuracy is maintained, but processing time increases and latency is reduced
Solution Approach 1:
The patent segments the object detection process into two distinct stages: (1) a first neural network stage that processes sensor data to generate object scores for multiple locations simultaneously, and (2) a second neural network stage that processes selected locations in batches to determine object properties. This segmentation allows the system to maintain accuracy while reducing latency by processing multiple locations in parallel during the first stage and then batching operations during the second stage.
Solution Approach 2:
The first neural network performs preliminary processing by generating object scores for all potential locations in a single forward pass before the second stage begins. This preliminary action identifies and ranks potential object locations, allowing the system to focus computational resources on the most promising candidates in subsequent stages, thereby reducing overall processing time while maintaining detection accuracy.
2Productivity
If the system uses multiple neural network stages with batch processing, then computational efficiency improves and latency reduces, but system complexity increases
Solution Approach 1:
The complex detection task is segmented into two specialized neural networks: the first network is optimized for generating object scores across multiple locations simultaneously, while the second network is optimized for determining object properties in batches. This segmentation of functionality allows each network to be relatively simple and efficient, avoiding the need for a single overly complex network, thereby improving computational efficiency while managing overall system complexity.
Solution Approach 2:
The two-stage neural network system provides multi-functionality: the first stage handles location scoring and selection, while the second stage handles property determination. This universal approach allows the system to perform multiple detection tasks (detection, selection, and property analysis) using a coordinated pair of networks rather than requiring separate specialized systems, improving efficiency while keeping individual network components manageable in complexity.
3Measurement precision
If the system selects locations iteratively one at a time, then object properties can be determined accurately, but processing speed decreases and latency increases
Solution Approach 1:
The patent segments the location processing into two phases: Phase 1 uses the first neural network to evaluate all locations simultaneously and generate object scores in parallel, identifying multiple candidate locations at once. Phase 2 then processes these selected locations in batches through the second neural network to determine properties. This segmentation eliminates sequential one-at-a-time processing while maintaining accuracy through the batch processing in Phase 2.
Solution Approach 2:
The first neural network performs preliminary selection of multiple candidate locations simultaneously before the property determination stage begins. This preliminary action creates a shortlist of high-probability object locations that are then processed in batches, allowing the system to maintain accurate property determination while significantly improving processing speed compared to iterative single-location processing.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for selecting locations in an environment of a vehicle where objects are likely centered and determining properties of those objects. One of the methods includes receiving an input characterizing an environment external to a vehicle. For each of a plurality of locations in the environment, a respective first object score that represents a likelihood that a center of an object is located at the location is determined. Based on the first object scores, one or more locations from the plurality of locations are selected as locations in the environment at which respective objects are likely centered. Object properties of the objects that are likely centered at the selected locations are also determined.