Obstacle Track Prediction for Accurate Autonomous Collision Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional collision detection methods in autonomous driving vehicles suffer from low accuracy, leading to frequent sudden braking and compromised safety due to geometric position overlapping and Kalman filter limitations.
Innovation Solution
A collision detection method that involves acquiring and analyzing video frames to determine the historical moving track of obstacles, using a pre-trained track prediction model to forecast potential paths, and modifying the vehicle's planned track based on predicted obstacle movements to prevent collisions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional geometric position overlapping or Kalman filter methods are used for collision detection, then the detection process is simple, but the accuracy is low leading to frequent sudden braking
Solution Approach 1:
The system performs preliminary actions by acquiring multiple video frames containing the obstacle target within a preset time period and analyzing them to obtain the historical moving track before collision detection. This preliminary analysis of historical movement patterns enables more accurate prediction of future positions, improving collision detection accuracy without requiring overly complex real-time computation during the detection phase itself.
Solution Approach 2:
The system applies dynamics by using a pre-trained track prediction model to dynamically predict the obstacle's future moving tracks based on historical data. The prediction model adapts to the obstacle's movement patterns and generates multiple possible future trajectories, allowing the collision detection system to account for dynamic changes in obstacle behavior rather than relying on static geometric overlap methods.
2Reliability
If traditional collision detection methods are used, then the system response is fast, but the safety is compromised due to low accuracy
Solution Approach 1:
The system performs preliminary action by pre-training the track prediction model offline and preparing the historical moving track analysis before actual collision detection is needed. This preprocessing enables the system to make rapid predictions during runtime by leveraging pre-computed patterns from multiple video frames, thus improving safety without incurring excessive processing delays during critical detection moments.
Solution Approach 2:
The system uses copying by creating multiple predicted moving tracks based on the historical moving track pattern. Instead of directly computing complex predictions in real-time, the system generates multiple possible future trajectories by copying and extrapolating historical movement patterns through the pre-trained model, enabling fast and reliable collision assessment.
3Measurement precision
If multiple video frames are analyzed to determine historical moving track, then the accuracy of track prediction is improved, but the computational complexity increases
Solution Approach 1:
The system applies preliminary action by performing the computationally intensive work of analyzing multiple video frames and training the prediction model in advance, before actual collision detection is required. This offline preprocessing shifts the computational burden to non-critical periods, allowing accurate track prediction during runtime with minimal real-time power consumption.
Solution Approach 2:
The system uses copying by storing the pre-trained prediction model and using it to generate multiple predicted tracks through pattern copying rather than re-computing from scratch. This approach replicates the successful prediction logic across multiple scenarios, maintaining high accuracy while reducing repeated computational overhead.
Data Source
AI summary
A collision detection method and device, as well as an electronic device and a storage medium are provided. The collision detection method includes: acquiring a plurality of video frames containing an obstacle target within a preset time period; analyzing the plurality of video frames to acquire a historical moving track of the obstacle target within the preset time period; inputting the historical moving track into a pre-trained track prediction model, to acquire at least one predicted moving track of the obstacle target; and performing a collision detection with a planned track of an autonomous vehicle and the at least one predicted moving track. The accuracy of collision detection can be improved.