Real-Time Object Detection With Training-Matched Decompression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Training machine learning models for object detection using encoded and decoded images leads to inaccurate real-time detection due to differences in image quality between training and real-time data, causing blurry edges and performance discrepancies.
Innovation Solution
Process images for real-time object detection using the same encoding and decoding techniques as the training process, ensuring the model is trained on similarly processed images to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If images are encoded and decoded to reduce bandwidth usage and cost, then transmission efficiency is improved, but image quality deteriorates causing blurry edges and inaccurate object detection
Solution Approach 1:
The system performs encoding before transmission and decoding after reception as preliminary actions to prepare images for efficient transmission while maintaining quality through controlled decompression parameters
Solution Approach 2:
The system changes compression parameters such as quality level, bitrate, or resolution during encoding and decoding to optimize the balance between bandwidth efficiency and image quality preservation, allowing adjustable trade-offs based on transmission conditions
2Productivity
If images are encoded and decoded for training, then data transmission efficiency is improved, but detection accuracy deteriorates due to edge bluriness
Solution Approach 1:
The system uses feedback mechanisms where decoded images are evaluated for quality and detection performance, and compression parameters are adjusted based on this feedback to maintain accurate object detection while preserving transmission efficiency
Solution Approach 2:
Images are encoded and decoded as preliminary processing steps before training and detection, with the decoding process carefully controlled to preserve edge sharpness and image quality necessary for accurate machine learning model training
3Quantity of substance
If multiple vehicles transmit images to cloud server, then training data quantity and diversity are improved, but processing complexity and cost increase
Solution Approach 1:
The system segments the image processing task by having each vehicle perform local encoding before transmission, reducing the processing burden on the cloud server and enabling efficient handling of large volumes of data from multiple vehicles
Solution Approach 2:
The cloud server performs universal decoding and training functions that can handle images from multiple vehicles simultaneously, using standardized processing pipelines that scale efficiently with the number of contributing vehicles
Data Source
AI summary
A system and method for deploying machine learning models with consistent fixed-point arithmetic processing. A computing device in a vehicle receives a trained machine learning model that was trained using training images processed with fixed-point arithmetic operations. The computing device receives images from a camera and processes the received images using fixed-point arithmetic operations that are consistent with those used during training. The machine learning model is executed using the processed images to detect objects such as vehicles, pedestrians, persons on bikes, stop lights, or stop signs. The computing device may comprise specialized hardware including a graphics processing unit (GPU), digital signal processing hardware decoder, or single-purpose hardware decoder such as an application-specific integrated circuit (ASIC) or field-programmable gate array (FPGA). The system may generate alerts or determine vehicle operation changes based on detected objects, and may operate in surveillance mode when the vehicle is parked.