Off-Road Object Tracking Using Terrain-Corrected Motion Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection and classification systems struggle in off-road environments due to invalid assumptions about continuous surfaces and progressive vehicle movement, leading to difficulties in establishing valid targets and accurate classification.
Innovation Solution
A target object tracking system that analyzes image data from sensors on a vehicle to identify image components, determine movement vectors, classify components as target or non-target, and modify target movement vectors based on non-target components to improve tracking accuracy over rough surfaces.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If object detection systems use assumptions about continuous surfaces and progressive vehicle movement, then detection accuracy is improved for on-road conditions, but the system becomes unreliable for off-road environments with irregular surfaces
Solution Approach 1:
The system dynamically adjusts its operation based on detected surface characteristics. When irregular surfaces are detected, the system switches from assuming progressive movement to allowing for sharp displacements, making the detection algorithm adaptive to different terrain conditions rather than relying on fixed assumptions
Solution Approach 2:
The system changes key parameters of the detection algorithm based on environment type. For off-road conditions, it modifies movement prediction parameters to account for non-progressive vehicle displacement, and adjusts classification thresholds to accommodate terrain-induced image variations
2Reliability
If object classification systems assume tight horizon envelopes for target vehicles, then false positives are reduced for on-road driving, but valid targets are missed in off-road conditions with elevation changes
Solution Approach 1:
The horizon envelope parameters are made dynamic rather than static. The system adjusts the acceptable range of vehicle positions in the image based on detected terrain characteristics, allowing the classification boundaries to expand or contract according to whether the vehicle is on a flat road or traversing uneven off-road terrain
Solution Approach 2:
The system performs preliminary terrain analysis using inertial sensors and image data before applying object classification. This preliminary detection of surface irregularities allows the system to pre-adjust classification parameters, preparing the algorithm to correctly identify targets even when they fall outside the standard tight horizon envelope
3Measurement precision
If the system corrects for vehicle movements caused by surface irregularities, then target tracking accuracy is improved, but the complexity of the tracking algorithm increases
Solution Approach 1:
The system introduces an intermediary correction step that uses inertial sensor data to compensate for vehicle movements in the image processing pipeline. Rather than completely redesigning the tracking algorithm, it applies correction vectors derived from accelerometer and gyroscope measurements to adjust target positions, effectively decoupling the complexity of movement compensation from the core tracking logic
Data Source
AI summary
A target object tracking system (1) includes a processor (5) for receiving image data (S1) captured by one or more sensor (7) disposed on the vehicle (2). The processor (5) is configured to analyse the image data to identify image components (IMC(n)) and to determine a movement vector (V(n)) of each image component (IMC(n)). The movement vectors each include a magnitude and a direction. At least one of the image components (IMC(n)) is classified as a target image component relating to the target object and at least one of the remaining image components (IMC(n)) as a non-target image component. The movement vector (V(n)) of the at least one target image component is modified in dependence on the movement vector of the or each non-target image component. The target object (3) is tracked in dependence on the modified movement vector of the at least one target image component. The disclosure also relates to a method and a non-transitory computer-readable medium.