Deep Neural Network Functional Module Deduplication for Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated vision processing systems require significant computational resources and memory to detect multiple objects, such as humans and vehicles, in a single image, leading to inefficiencies and increased costs.
Innovation Solution
Implementing a deep neural network functional module deduplication method that uses a single AI model to detect both human and vehicle objects in an image, reducing the need for multiple models and memory copies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple separate detection neural networks are used to detect different object types (humans, vehicles), then detection accuracy for each object type is maintained, but memory usage and computational power requirements increase significantly
Solution Approach 1:
The patent merges multiple separate detection neural networks into a single unified detection network. The network architecture integrates multiple detection heads that can detect different object types (humans, vehicles, etc.) simultaneously within one model, eliminating the need for multiple separate models and their associated memory requirements.
Solution Approach 2:
The unified detection network is designed with multi-functionality to handle various object detection tasks. The network contains multiple detection heads, each specialized for different object types, allowing a single model to perform multiple detection functions that previously required separate specialized models.
2Adaptability or versatility
If multiple separate detection neural networks are used to detect different object types, then comprehensive object detection is achieved, but processing speed decreases due to sequential processing requirements
Solution Approach 1:
By combining multiple detection networks into one unified architecture, the system enables parallel processing of different object types within a single forward pass through the network, eliminating the sequential processing bottleneck that reduced productivity.
Solution Approach 2:
The unified detection network is segmented into multiple independent detection heads, each responsible for detecting specific object types. This segmentation allows each head to process its target objects independently and simultaneously, improving overall processing speed while maintaining comprehensive detection coverage.
3Adaptability or versatility
If separate functional modules are programmed for each application (human detection, vehicle detection), then specialized detection functionality is achieved, but system complexity and processing overhead increase
Solution Approach 1:
The unified detection network provides specialized detection capabilities for different object types within a single modular architecture. Each detection head is specialized for its object type (e.g., human detection head, vehicle detection head), but all heads share the same backbone network and can be managed through a single system interface, reducing overall system complexity.
Data Source
AI summary
In one embodiment, an image sensor captures an image and passes the image to an image processing system that detects one or more vehicle objects and one or more human objects, using a single detection function having a single artificial intelligence (AI) model. Each of the detected objects correspond to a portion of the captured image. For each vehicle object, the corresponding image portion is processed to determine a plurality of properties of the vehicle, including, e.g., type (truck, van, sedan, motorcycle, etc.), color, and license number. For each human object, the corresponding portion of the image is processed to determine a plurality of properties of the human, e.g., gender, approximate age, skin tone, eye color, hair color, and facial landmarks for facial recognition. The captured image is annotated to describe the objects and their properties, and annotated image is transmitted to a service for use by an application.