Vehicle Occupancy Prediction Under Camera Occlusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing digital imaging systems in vehicles face challenges with occlusions that lead to inaccurate occupancy class predictions, particularly in self-driving cars, where prolonged occlusions can result in unreliable seat occupancy classification.
Innovation Solution
A computer-implemented method for occupancy class prediction and occlusion value determination that employs a cascade structure with multiple modes (regular, short-term, long-term, and HMI occlusion) to handle varying occlusion levels, using weight masks and time-fused predictions to stabilize occupancy class determination, and involves a human-machine interface for assistance during prolonged occlusions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a camera is used to monitor the vehicle interior, then additional sensing possibilities are offered, but inaccurate observing results occur when occlusions are present
Solution Approach 1:
The system dynamically switches between different occupancy detection modes (occlusion mode, ambiguous mode, non-occlusion mode) based on real-time occlusion detection. This dynamic adaptation allows the system to maintain reliable occupancy classification by adjusting its detection strategy according to the current occlusion state, resolving the contradiction between versatility and reliability.
Solution Approach 2:
The system changes operational parameters by switching between different detection modes with distinct characteristics. In non-occlusion mode, standard detection parameters are used, while in occlusion modes, modified parameters and alternative detection strategies are applied. This parameter adaptation enables the system to maintain accuracy across varying occlusion conditions.
2Adaptability or versatility
If traditional sensors are replaced by cabin view cameras, then a wide range of additional possibilities are offered, but observation becomes impossible during prolonged occlusions
Solution Approach 1:
The system introduces an intermediary occlusion detection mechanism that identifies when occlusions occur and triggers alternative detection modes. This intermediary layer prevents complete loss of observation capability by activating backup strategies (such as using different image regions or temporal patterns) when the primary view is blocked.
Solution Approach 2:
The system performs preliminary occlusion detection and mode switching before complete observation failure occurs. By detecting occlusion conditions early and transitioning to alternative modes proactively, the system maintains observation capability throughout the occlusion event rather than failing completely.
3Reliability
If occlusion detection is implemented, then unreliable classification during occlusion is addressed, but system complexity increases
Solution Approach 1:
The system segments the occupancy detection process into distinct operational modes (occlusion mode, ambiguous mode, non-occlusion mode), each with specialized detection logic. This segmentation allows the system to apply simpler, mode-specific algorithms rather than requiring a single complex algorithm to handle all conditions, thereby managing complexity while maintaining reliability.
Solution Approach 2:
The occlusion detection mechanism serves multiple functions: it identifies occlusion conditions, triggers mode transitions, and informs the occupancy classification process. This multi-functionality reduces the need for separate dedicated components for each function, thereby limiting the increase in system complexity while achieving reliable classification.
4Reliability
If mode switching based on occlusion values is implemented, then occupancy class prediction reliability is enhanced, but processing time increases
Solution Approach 1:
The system applies partial processing by focusing computational resources on the currently active mode's specific requirements. Rather than executing all detection algorithms in every frame, the system performs only the necessary processing for the current occlusion state, reducing overall processing time while maintaining prediction reliability through targeted analysis.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computer implemented method for occupancy class prediction, preferably for occupancy class detection in a vehicle, may include the following steps carried out by computer hardware components: determining for a plurality of points of time measurement data related to an area; determining for a plurality of points of time occlusion values based on the measurement data; selecting for a present point of time one of a plurality of modes for occupancy class prediction based on the occlusion values for at least one of the present point of time and a previous point of time and/or based on one of the plurality of modes for occupancy class prediction selected for the previous point of time; and determining for the present point of time one of a plurality of predetermined occupancy classes of the area based on the selected mode for the present point of time.