Vehicle Camera Collision Detection Using Neural Time-to-Collision

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Identifying the distance of objects to a vehicle requires significant memory, time, and computing resources, which can be improved.

Innovation Solution

A neural network processes images captured from a vehicle's camera at different times to determine a time to collision (TTC) by scaling and segmenting objects based on their movement relative to the vehicle, generating a binary segmentation map to identify potential collisions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional methods are used to identify object distances, then measurement precision is achieved, but computing resources and time consumption increase significantly

Engineering Contradiction:

Improveobject distance identification accuracyVSAvoidcomputing resource efficiency

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The patent replaces traditional mechanical/computational methods of distance measurement with a neural network-based optical system. The neural network processes images from a single camera to directly estimate distances and time-to-collision values, substituting complex computational algorithms with a trained neural model that operates more efficiently while maintaining measurement precision.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent transforms the problem from measuring absolute distances to estimating time-to-collision parameters. By changing the output parameter from distance metrics to time-based collision predictions, the system achieves the required measurement precision with reduced computing resources, as the neural network directly regresses time-to-collision values from image sequences.

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If traditional optical flow algorithms are used for object detection, then movement analysis accuracy is improved, but device complexity and processing time increase

Engineering Contradiction:

Improveobject movement analysis accuracyVSAvoidalgorithm complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent substitutes traditional optical flow algorithms with a neural network-based approach. Instead of using complex mathematical computations to track pixel movements, the neural network directly processes image sequences to identify objects and estimate their motion, thereby reducing algorithmic complexity while maintaining movement analysis accuracy.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent uses a neural network model that has been trained to replicate the functionality of traditional optical flow algorithms. The neural network learns to copy the movement detection capabilities of optical flow methods through training on labeled data, achieving similar accuracy with reduced computational complexity during inference.

Inventive Principle:

Principle #26Copying

Data Source

PatentUS20260054723A1Object detection and collision avoidance using a neural network

Publication Date: 2026.02.26 NVIDIA CORP

AI summary

Apparatuses, systems, and techniques to identify objects in view of a camera associated with a vehicle. In at least one embodiment, objects with which a vehicle may collide are identified, based on, for example, a difference between a size of an image of the objects detected at a first point in time and a size of an image of the objects detected at a subsequent point in time.