Contact person: Business Client

Phone:Show

Email:Show

Location: New Delhi, India

Budget: Recommended by industry experts

Time to start: As soon as possible

Project description:
"I’m looking for a streamlined deep-learning solution that can spot pedestrians, vehicles, and road signs from an in-vehicle camera while running on a constrained edge processor. My top priority is low power consumption, so every design choice—from architecture to post-training optimisation—needs to favour efficiency without letting accuracy slide.

Here’s what I need, kept intentionally lean so we stay within scope:

• A compact object-detection model (e.g., MobileNet- or Nano-YOLO-class) trained or fine-tuned to recognise the three target classes.
• Conversion and optimisation for common automotive edge targets (TensorRT, ONNX, or similar), with clear instructions so I can reproduce the build.
• A short inference script that takes a video stream and returns bounding boxes plus class labels in real time.
• A brief report (bullet points are fine) outlining model size, FPS on a reference device, and measured power draw or estimated watt-usage.

Speed is important—please be ready to turn around an initial working build ASAP. If additional data or hardware details are required, flag them early so we can keep momentum.

Additional information:

While VANET-DDoSNet++ shows high detection accuracy coupled with very strong mitigation capabilities, it faces certain limitations inherent to its design. The multi-layered architecture, composed of complex deep learning models and blockchain-based reporting, brings computational overhead and energy expenses, making it an uphill task to be deployed on resource-limited vehicular devices. Additionally, the system may suffer from increased false positives or false negatives under highly dynamic network conditions, where node mobility and topology changes are rapid and unpredictable. These issues may impact real-time threat mitigation and compromise network stability.

Scalability also remains a concern—especially in maintaining blockchain consensus during high vehicular density or when operating across large-scale networks. Moreover, the gap between simulation environments and real-world VANET scenarios introduces transferability challenges, where results obtained in controlled settings may not fully reflect practical performance.

To overcome these limitations, future research can explore the following avenues:

Development of lightweight deep learning models suitable for edge deployment on in-vehicle processors without compromising detection accuracy.

Optimization of blockchain mechanisms using energy-efficient and low-latency consensus protocols tailored for vehicular environments.

Deployment and testing on actual vehicular platforms to validate system performance under real-world latency, bandwidth, and energy constraints.

Incorporation of online learning and adaptive thresholding to dynamically tune detection parameters in response to evolving attack strategies.

Expansion of the attack detection framework to handle hybrid, multi-vector, and zero-day threats using few-shot or meta-learning techniques.

Integration of explainable AI to enhance decision transparency and trust among vehicle manufacturers and users." (client-provided description)