Sparse Object Proposals for ADAS Using Tracked Detections
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern object detectors in video sequences rely on computationally heavy image representations and ignore temporal correlation, leading to inefficient object detection in applications like autonomous driving and surveillance.
Innovation Solution
The system generates sparse object proposals by combining image-based proposals from the current frame with tracked bounding boxes from previous frames, using a novel module that learns to propose new candidates only when necessary, such as when new objects appear or the tracker fails, thereby reducing the number of proposals evaluated by the detector.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If modern object detectors use strong image representations and category independent object proposal mechanisms, then detection accuracy is improved, but computational complexity increases
Solution Approach 1:
The system performs object tracking in advance to generate tracked detections from previous frames, which are then used as informed proposals for the current frame. This preliminary tracking action reduces the search space for the object detector, allowing it to focus computational resources on fewer, more promising candidate regions while maintaining high detection accuracy
Solution Approach 2:
The detection process is segmented into two distinct phases: (1) tracking-based proposal generation that identifies candidate regions using temporal information, and (2) detector-based classification that applies strong image representations only to these segmented candidate regions. This segmentation avoids applying computationally heavy detection algorithms to the entire image
2Reliability
If object detectors are applied independently in every frame, then detection reliability is improved, but temporal correlation is lost and computational efficiency decreases
Solution Approach 1:
The system maintains continuous tracking of detected objects across video frames, creating an unbroken chain of temporal association. This continuous tracking preserves temporal correlation while allowing the detector to operate less frequently (only on frames where tracking is uncertain or new objects appear), thereby improving computational efficiency without sacrificing detection reliability
Solution Approach 2:
The system uses feedback from tracking results to guide subsequent detection operations. Tracked detections from previous frames provide feedback that informs where to search for objects in the current frame, allowing the detector to focus on relevant regions and reducing the need for full-frame independent detection while maintaining reliability through temporal consistency
3Measurement precision
If the number of object proposals is increased, then recall is improved, but computational overhead increases
Solution Approach 1:
Tracking operations are performed in advance to generate a refined set of candidate proposals before the object detector processes them. This preliminary tracking action filters out unlikely candidate regions and concentrates computational resources on a smaller, more promising set of proposals, achieving high recall without proportionally increasing computational overhead
Solution Approach 2:
The system applies different processing qualities to different regions: tracked detections with high confidence receive minimal processing (direct use as proposals), while regions with low confidence or new objects receive full detector processing. This local quality adjustment ensures high recall for important regions while reducing overall computational overhead by avoiding uniform high-cost processing across all regions
Data Source
AI summary
A system and method are provided for driving assistance. The system includes an image capture device configured to capture a video sequence, relative to an outward view from a vehicle, which includes a set of objects and is formed from a set of image frames. The system includes a processor configured to detect the objects to form a set of object detections, and track the set of object detections over the frames to form tracked detections. The processor is configured to generate for a current frame, responsive to conditions, a set of sparse object proposals for a current location of an object based on: (i) the tracked detections of the object from an immediately previous frame; and (ii) detection proposals for the object derived from the current frame. The processor is configured to perform an action to mitigate a likelihood of potential harmful due to a current object location.